# Built with Opus 4.6: a Claude Code hackathon: Project Gallery - **Event:** [Built with Opus 4.6: a Claude Code hackathon](https://cerebralvalley.ai/e/claude-code-hackathon) - **When:** Feb 10 at 12:00 PM – Feb 17 at 10:00 AM (EST) - **Where:** Location TBA - **Hosts:** [Cerebral Valley](https://cerebralvalley.ai/u/cv), [Anthropic](https://cerebralvalley.ai/u/anthropicai), [Joshua Brown](https://cerebralvalley.ai/u/joshuabrown) - **Projects:** 227 (8 placed) - **Page:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery ## Projects ### 1. Tomas Forge is a self-extending Claude Code agent that never says "I can't do that" — instead, it builds the capability on the spot. When Forge encounters a request it can't handle, it autonomously designs, generates, validates, and registers a new MCP tool server, then uses it immediately to answer the original question. Every tool it builds is a real FastMCP server that persists forever, so the agent gets permanently smarter with every interaction. What sets it apart is rigor: every generated server passes through an 11-check AST-based validator before it goes live, and a built-in security policy prevents it from building anything dangerous. Forge uses five Claude Code platform features working together — Custom Agents, MCP Servers, Skills, Hooks, and Opus 4.6's extended thinking — to create an agent that doesn't just use tools, but manufactures them. - **Team:** [Tomas Mihalyi](https://cerebralvalley.ai/u/tomyister) - **GitHub:** https://github.com/tomasmihalyi/claude-forge - **Demo video:** https://youtu.be/wX99Kd6hGYQ - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=1 ### 2. FirstClaim Medical billing errors cost US providers over $125B annually. Doctors' notes must be translated into standardized codes (ICD-10, CPT) to get paid. Wrong code, missed modifier, or impossible pairing means a denied claim. FirstClaim is an AI-native billing application powered by Claude Opus 4.6. Paste clinical notes and a validation pipeline extracts diagnoses and procedures, matches them against a 72,000-entry ICD-10 database, runs compliance checks (age/sex rules, code compatibility, medical necessity), flags errors with dollar amounts at risk, and calculates a risk score. It caught a mammogram billed for a 30-year-old male before it reached an insurer. The claim workspace makes every finding actionable. Fix issues through conversation: "remove that line item," "swap this code," "bill at a lower level." The claim updates, findings resolve, and risk recalculates in real time. Session history lets you pick up where you left off. Cleaner claims, better reimbursement, the first time. - **Team:** [Tyler Gibbs](https://cerebralvalley.ai/u/tylergibbs) - **GitHub:** https://github.com/tylergibbs1/firstclaim - **Demo video:** https://youtu.be/_grR_Zve1D4 - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=2 ### 3. Respiro Respiro is a macOS menu bar app that detects stress from your screen and guides you through micro-practices. It captures your screen and sends it to Opus 4.6 Vision API to read stress signals — tab overload, error dialogs, back-to-back meetings. The core challenge: knowing when NOT to help. Opus uses extended thinking to reason about whether to intervene or stay silent. When stress peaks, Opus calls tools (practice catalog, user history) to select the best exercise — a 2-minute breathing session from the menu bar. Opus 4.6 powers every layer: • Vision API — multimodal stress detection • Extended Thinking — reasoning about when to stay quiet • Tool Use — practice selection from history • 1M Context — end-of-day reflection We built AI playtesting: hundreds of stress scenarios trained the engine, catching 150+ bugs. Zero dependencies, pure Swift 6, privacy-first. - **Team:** [Kostiantyn Vlasenko](https://cerebralvalley.ai/u/KostyaV) - **GitHub:** https://github.com/kv100/respiro-macos - **Demo video:** https://www.youtube.com/watch?v=CVSdXwF2pX4 - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=3 ### 4. Replace SaaS - Build It Yourself My wife owns her own business as a tattoo artist. She pays 200 a month for booking software to communicate with her clients, and yet that booking software isn't quite what she'd like. Today AI models like Opus 4.6 are so good that I wanted to see if I could build this app myself for her. If so, I'd be able to replace the 200 a month service that she pays for while also giving her and her clients a truly custom experience. - **Team:** [Brian Rogers](https://cerebralvalley.ai/u/brogers) - **GitHub:** https://github.com/woweow/booking-app - **Demo video:** https://youtu.be/jhAxF58AcMs - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=4 ### 5. Nanas Sound Conductr is a browser-based MIDI instrument where you play and Claude Opus 4.6 directs the band. Four tracks of drums, bass, melody, and harmony generate in real time through a C engine compiled to WebAssembly, outputting MIDI to any DAW or synth. The system operates on three timescales. The C engine generates notes every 15ms using Euclidean rhythms, walking bass, and random-walk melodies. A performance analyzer extracts features from your playing. Opus 4.6 receives that summary and reshapes the arrangement. The engine never waits for the AI, so latency is musically invisible. Speak or type commands like "make it funky" and Opus translates musical intent into parameter changes. A fallback director keeps the instrument working without the API. About 4,800 lines of code, no frameworks. - **Placement:** Finalist - **Team:** [Asep Bagja Priandana](https://cerebralvalley.ai/u/nanas) - **GitHub:** https://github.com/nanassound/conductr - **Demo video:** https://www.youtube.com/watch?v=X6CqJoyj0kI - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=5 ### 6. BrandtDev Try it live: https://forge.brandt-thomassen.dk - Click "Try the Demo" on the login page. Pre-loaded with contacts, segments, templates, and flows. AI features and email sending are fully functional. Email marketing made effortless for small businesses. Today's tools are either enterprise complexity or bare-bones APIs. Forge fills the gap: guided flows, AI-native integration, and a visual email builder - signup to sending in minutes. Core features: drag-and-drop email builder, AI assistant via Sonnet 4.5 (subject lines, email bodies, improvements, alt text), visual flow designer for automations, dynamic segments, team management, and GDPR-ready consent system. Built from create-next-app to production in 5 days with Opus 4.6. 35 commits, solo developer. Workflow: CLAUDE.md (architecture), PROJECT_PLAN.md (tracking), LEARNINGS.md (debugging knowledge). 22 Supabase tables with full RLS. Stack: Next.js 16, React 19, Supabase, shadcn/ui, Resend, React Email, dnd-kit, Vercel. - **Team:** [Christian Thomassen](https://cerebralvalley.ai/u/brandtdev) - **GitHub:** https://github.com/BrandtChristian/forge/ - **Demo video:** https://youtu.be/v99pd_MIagk - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=6 ### 7. Seeder Works DraftCut is an AI-powered video editor that runs entirely in your browser, turning text prompts into professional videos with motion graphics, animations, and effects. Upload your images or video clips, describe what you want, and AI generates a complete timeline with Ken Burns effects, crossfade transitions, animated text reveals, intro/outro cards, and lower thirds—eliminating hours of manual editing work. The agentic refinement system iteratively analyzes and improves video quality like "Claude Code for videos." All processing happens locally with 100% privacy (assets never leave your device), and exports YouTube- ready WebM videos directly in the browser. DraftCut makes professional video production effortless for anyone, whether creating product showcases, social media content, or image slideshows—no expertise required. - **Team:** [Dhonam Pemba](https://cerebralvalley.ai/u/dhonam) - **GitHub:** https://github.com/seeder-works-studio/DraftCut - **Demo video:** https://www.youtube.com/watch?v=wwvE3E7UNtM - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=7 ### 8. Routine 25 Most mornings are lost before they start — hitting snooze, rushing out the door, feeling behind from the moment you wake up. But with a little planning, you can accomplish more in your first hour than you ever thought possible. Routine 25 makes that happen. The app helps anyone set up a routine, even if they've never had one. Pick activities with or without your AI coach and build your perfect morning. Based on your activities and when you need to finish, your wake-up time is automatically calculated. Have high-priority items? Pin an activity to a specific time so it stays locked in, even if the rest of your routine changes. Built-in focus features keep you on track. Block distracting apps while a routine is active, and track progress via Live Activities on your lock screen. Need help adding activities? Want to analyze your sleep and fitness? Need a guided meditation or to log a meal? The AI coach handles it all through a simple conversation. Reclaim your mornings with Routine 25. - **Team:** [Jason Ellington](https://cerebralvalley.ai/u/jasone) - **GitHub:** https://github.com/jasonellington/r25-app - **Demo video:** https://youtu.be/GBRQc2d7SL4 - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=8 ### 9. Techulus A fast, offline security scanner that protects AI agents from prompt injection, credential leaks, and code injection attacks, the security layer every AI agent needs. Since AI blindly trust most inputs they receive, Arx is the immune system they're missing. - **Team:** [Arjun Komath](https://cerebralvalley.ai/u/arjunkomath) - **GitHub:** https://github.com/arjunkomath/arx - **Demo video:** https://youtu.be/ogP5NoS6BlQ - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=9 ### 10. PRISMA PRISMA makes Monte Carlo simulation accessible to anyone with a spreadsheet. Upload messy ops data, get instant dashboards and insights, then simulate decisions before you commit. Ask "What if I cut from 7 to 5 drivers?" - 1,000 simulations run, showing probability distributions and sensitivity analysis. Verdict: don't cut (34% positive). Follow up: "Keep 6, remove the weakest?" - PRISMA identifies the problem driver, simulates again: 63% positive. Not just answers - better questions. Built with Claude Code and Opus 4.6. The full reasoning chain: understanding messy data, generating structured tool_call specs, formulating simulation parameters, translating probabilistic outputs into business advice. ~12K lines vanilla JS. Client-side - no data leaves your machine. Vercel deployed. Built by an ops leader who scaled teams across Big Tech and high-growth startups. - **Team:** [Muzaffer Hizel](https://cerebralvalley.ai/u/Muz) - **GitHub:** https://github.com/MuzafferH/prisma-decision-engine - **Demo video:** https://www.youtube.com/watch?v=wB3x6xUImqQ - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=10 ### 11. Jason Tangen ClassBuild turns a topic description into a complete, multimedia course in minutes — structured syllabus, research with validated citations, interactive readings, practice quizzes, in-class assessments, discussion prompts, activities, narrated audio, slide decks, and AI infographics. Every output is shaped by five evidence-based learning principles: retrieval practice, interleaving, dual coding, concrete examples, and spacing. Opus 4.6 powers a five-stage pipeline — curriculum design, learning principle mapping, research synthesis, content generation, and assessment — where each stage feeds into the next. The science isn't applied as a finishing pass; it shapes the course from the first step. Runs entirely in the browser, no backend. Any subject-matter expert can produce pedagogically sound materials for any topic, any audience. - **Team:** [Jason Tangen](https://cerebralvalley.ai/u/tangenjm) - **GitHub:** https://github.com/jtangen/classbuild - **Demo video:** https://youtu.be/3_uXefhkGWQ - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=11 ### 12. Nexus Nexus is a full-stack platform that transforms how scientists run experimental optimization campaigns. Instead of manually choosing algorithms, tuning hyperparameters, and guessing when to switch strategies, Nexus autonomously diagnoses, decides, and explains — acting as a scientific reasoning partner, not a black box. Scientists spend 80% of optimization time on algorithm selection and debugging, not science. Existing tools like Optuna or Ax give you the next point to evaluate — but they don't tell you *why*, they can't detect when your instrument has drifted, and they start from scratch every campaign. Nexus tells you WHY, not just WHAT's next. - **Team:** [Sissi Feng](https://cerebralvalley.ai/u/SissiF) - **GitHub:** https://github.com/SissiFeng/optimization-copilot - **Demo video:** https://www.youtube.com/watch?v=XMOj8Qud7jk - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=12 ### 13. AiBrush fcoreutils: Claude Code autonomously rewrites GNU coreutils in Rust - up to 54x faster, 100% compatible. We gave Claude Code a small Linux machine with full sudo access, defined one goal (rewrite GNU coreutils with 100% compatibility and 10x-30x performance), and walked away. Zero human-written code. Claude Code consumed the full $500 API credit in 3 days + another 3 days Claude Max 20 - autonomously designing architecture, implementing 10 SIMD-accelerated tools, publishing 62+ releases to crates.io, and passing 826/826 compatibility tests. Results (independently verified via GitHub Actions): - wc: 54.3x faster than GNU, 28.6x faster than uutils/coreutils - sort: 18.1x faster than GNU, 15.1x faster than uutils - uniq: 13.4x faster than GNU - base64: 8.8x faster than GNU - cut/tr/tac: 5-7x faster than GNU Key: Claude Code didn't just match uutils/coreutils - it surpassed it across every tool. A single autonomous AI agent outperforming a major community-maintained Rust project. - **Team:** [Tarek Abdellatef](https://cerebralvalley.ai/u/tarekbadrsh) - **GitHub:** https://github.com/AiBrush/coreutils-rs - **Demo video:** https://www.youtube.com/shorts/_Z-6CIKUZ80 - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=13 ### 14. Mufid Team SQLite is the most deployed database engine in the world -- yet it was locked behind C expertise, native toolchains, and platform-specific binaries. The knowledge to reimplement it -- B-trees, bytecode compilation, virtual machine execution -- was reserved for specialists. AI breaks that barrier. It takes something powerful locked behind expertise and infrastructure and puts it in everyone's hands. And yes, sqlite-purerb changes that. Built with Claude Code, we reimplemented SQLite's internals entirely in pure Ruby: the pager, B-tree, bytecode compiler, and VDBE. Claude Code made this feasible. It navigated SQLite's C source, translated low-level algorithms into idiomatic Ruby, and drove a test-driven workflow validated against the reference implementation. This project joins Ruby's "Rewrite C in Ruby" revolution. Pure Ruby means true portability -- no compiler needed. - **Team:** [Mufid Afif](https://cerebralvalley.ai/u/mufid) - **GitHub:** https://github.com/mufid/sqlite-purerb - **Demo video:** https://www.youtube.com/watch?v=PIF3URvsoZA - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=14 ### 15. VoxPolicy VoxPolicy is a reasoning-first multilingual voice agent for call centers showcasing Claude's transparent reasoning, autonomous tool use, and cross-lingual understanding. Key Features: - Automatic customer identification by phone number - Real-time voice (Deepgram STT + ElevenLabs TTS) - Visible AI reasoning — Claude's thinking displayed live - Multilingual — English, Spanish, Mandarin with auto-detection - Customer memory persists across interactions - Autonomous tool use — search policies, update memory, or escalate - Smart escalation with structured handoff summaries Tech Stack: React, FastAPI, Claude Sonnet with extended thinking, Deepgram, ElevenLabs, SQLite Why "Amplify human judgment": VoxPolicy doesn't replace agents — it makes them better. Complex situations escalate with full context: customer history, sentiment, churn risk, and recommended actions. Visible reasoning lets supervisors understand every AI decision, building trust in autonomous customer service. - **Team:** [Udam Liyanage](https://cerebralvalley.ai/u/udamliyanage) - **GitHub:** https://github.com/UdamLiyanage/vox-policy - **Demo video:** https://drive.google.com/file/d/1PU6-DTTwOMBFVrq1sZA2g0qpC32Vj2QP/view - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=15 ### 16. Zeppelin Zeppelin is a production-grade, S3-native vector search engine built entirely by Claude Opus 4.6 in long-running agentic sessions. 21,000 lines of Rust, 451 tests, 26 TLA+ formal proofs for a complete distributed systems architecture. Opus autonomously planned, implemented, debugged, and iterated for hours at a stretch, discovering invisible concurrency bugs through probabilistic reasoning (computing exact failure distributions across namespaces), writing TLA+ specs that caught bugs and races no test could find and uncovered framework-level precision bugs in Rust's serde library. In the end we have a working database that stores vectors in S3, scales horizontally with zero coordination with performance in the same order of magnitude as that of mature database companies. Check it out at https://zepdb.github.io/ I have hosted a Zeppelin instance with an embedding endpoint for everyone to play around with. Install it as a skill by doing: "curl -sL https://zepdb.github.io/install.sh | sh" - **Placement:** Finalist - **Team:** [Anup Ghatage](https://cerebralvalley.ai/u/ghatage) - **GitHub:** https://github.com/zepdb/zeppelin - **Demo video:** https://youtu.be/j3dGGl5R1jQ - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=16 ### 17. The SHIELD AI agents are deployed everywhere, but nobody's watching them. A misconfigured agent can exfiltrate data or bypass controls. SwarmShield firewall is like Cloudflare for AI Agents, it intercepts every action in real time through a sub-millisecond policy engine. When an event is flagged, SwarmShield spawns special Claude Opus 4.6 agents to deliberate. The agents - Security Analyst, Threat Hunter, Compliance Auditor, Risk Assessor - analyze in parallel, debate across rounds, argue and counter-argue & vote using configurable consensus. Every verdict includes confidence scores, reasoning & dissents. When handling sensitive data, agents operate in ephemeral sessions. They deliberate, reach a verdict, then all traces are cryptographically destroyed, only the verdict survives. GDPR-compliant by design. Built on Elixir/OTP (tech behind WhatsApp’s 2 billion users). Each deliberation runs as an isolated lightweight process. Millions run simultaneously. If one crashes, others don't notice. - **Team:** [Anyona Obutu](https://cerebralvalley.ai/u/Claudius) - **GitHub:** https://github.com/obutuz/swarmshield - **Demo video:** https://www.youtube.com/watch?v=Kr2qUQcs9nQ - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=17 ### 18. RootedAI KisanMind is an AI-powered agricultural intelligence system that answers every farmer's critical question: "What should I plant this season to make the most money?" By synthesizing data across 6 intelligence sources and leveraging Claude Opus 4.6's extended thinking, KisanMind delivers comprehensive, profit-optimized farming recommendations in plain language accessible to farmers with limited literacy and technology access. KisanMind breaks these barriers by providing: ✅ Voice-first interface in 4+ regional languages (Hindi, Marathi, Tamil, Telugu) ✅ Visual intelligence - Upload photos to detect soil type and crop diseases (89% & 93% accuracy) ✅ Audio-guided capture for low-literacy farmers ✅ Multi-agent orchestration synthesizing 6 intelligence sources in parallel ✅ Profit-optimized recommendations with exact numbers, not vague advice ✅ Month-by-month action plans with government scheme recommendation - **Team:** [Ganesh Khovalan](https://cerebralvalley.ai/u/living_cool) - **GitHub:** https://github.com/livingcool/kisanmind - **Demo video:** https://drive.google.com/drive/folders/1Xsp8Np2Ul62SZcqCx9K0D9urUwR-PwwJ?usp=sharing - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=18 ### 19. Louis Download Link for Clover v0.0.0 Test Clover with this direct download: https://github.com/jeremiah-louis/clover/releases/download/v0.0.0/Clover-0.0.0.dmg ​ macOS "App is Damaged" Fix The app lacks code signing, triggering macOS Gatekeeper warnings. Right-click the mounted app, select "Open," then confirm in the dialog—this adds a permanent exception Clover is a desktop AI tool that turns plain-English hardware ideas into working, simulated Arduino projects in seconds. You describe what you want to build, and Clover automatically generates the code, designs the circuit, compiles it, and runs it in a live visual simulator no physical components, setup, or prior hardware experience required - **Team:** [Jeremiah-louis ojie](https://cerebralvalley.ai/u/louisojie) - **GitHub:** https://github.com/jeremiah-louis/clover - **Demo video:** https://youtu.be/67h6yHwtII4 - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=19 ### 20. ClaudingAsh CLAUDETANGLE - https://claudetangle.vercel.app/ Quantum computing is transformative, yet understanding it remains locked behind dense textbooks, expensive lab access, and years of physics prerequisites. We believe everyone deserves to build quantum intuition, not just PhDs. Claudetangle is an interactive 3D quantum computing visualizer that makes quantum concepts tangible. Users watch qubits rotate on Bloch spheres, see entanglement form as glowing beams, and observe superposition collapse in real time, all in the browser, no installations required. We built five guided chapters from classical bits to entanglement, a step-by-step walkthrough that pauses at each gate with plain-English explanations, an interactive sandbox for free experimentation, and visualizations of real algorithms including Shor's factoring. What once required a physics course now takes an afternoon of exploration. Quantum literacy should be a playground, not a privilege. - **Team:** [Ashok Muralidharan](https://cerebralvalley.ai/u/claudingash) - **GitHub:** https://github.com/Ashok93/claudetangle - **Demo video:** https://www.youtube.com/watch?v=FqVOctJjEEE - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=20 ### 21. VenTech Case Nexus is an AI-powered caseload intelligence platform built for Georgia public defenders who manage hundreds of active cases. By loading a defender’s entire caseload into a million-token context window, the system identifies patterns that humans often miss, such as recurring unconstitutional stops by the same officer across unrelated files. The platform utilizes 60,000 tokens of extended thinking to flag speedy trial risks and constitutional violations. Through Cascade Intelligence, Claude acts as an autonomous investigator using nine specialized tools to pull case files, research statutes, and verify citations. To stress-test legal strategies, the system runs adversarial simulations by chaining three Claude sessions representing the prosecution, defense, and judge with over 80,000 tokens of reasoning. This 20,000-line project was built entirely with Claude Code, demonstrating high-capacity inputs and complex multi-phase reasoning. - **Team:** [Eli Benveniste](https://cerebralvalley.ai/u/ven) - **GitHub:** https://github.com/eliBenven/case-nexus - **Demo video:** https://youtu.be/r0dHdom1m8Q - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=21 ### 22. Midspiral As LLMs become more capable of writing code, the programmer's role shifts from implementation to describing intent. Two major open questions emerge in AI programming: 1) How do we know AI-generated code is correct, not just plausible? 2) How do we know the code accurately captures the programmer's intent? Our recent work addresses the first question by adding formal verification to the AI code generation loop — forcing the AI to mathematically prove its code is correct. But the second question remains: does the proved specification capture the intent? Claimcheck closes the gap between “the code is guaranteed correct” and “the code makes the intended guarantees.” We use the Claude API to conduct round-trip informalization: it translates formal language to English without knowledge of the initial input. It then compares the intent to the guarantee to ensure they match. On 108 comparisons, it got 96.3% correct in 2 API calls per run batched per file. - **Team:** [Fernanda Graciolli](https://cerebralvalley.ai/u/heyyfernanda), [Nada Amin](https://cerebralvalley.ai/u/namin) - **GitHub:** https://github.com/metareflection/claimcheck - **Demo video:** https://www.youtube.com/watch?v=Dwt-gkLiSYU - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=22 ### 23. CouncilSim $64B+ in data center projects blocked by local opposition. Petitioners enter council meetings unprepared, and officials make decisions based on heated, one-sided testimony. CouncilSim helps both sides. Upload a proposal PDF and watch five AI personas — moderator, petitioner, and concerned residents — debate live through opening statements, public comment, rebuttals, Q&A, and deliberation. Built with Claude Code and Opus 4.6, CouncilSim uses the Claude Agent SDK to orchestrate four agents with three custom MCP tools. One scans real Facebook groups for community sentiment. Another analyzes the full proposal. The debate streams via WebSocket in real-time. Petitioners get an approval score, opposition arguments with battle-tested rebuttals, and a strengths/weaknesses breakdown. Better-prepared petitioners mean better-informed officials — decisions grounded in data, not fear. Go from rejected to approved — with confidence. https://councilsim.live/ - **Team:** [Dorian Sadellari](https://cerebralvalley.ai/u/dorian) - **GitHub:** https://github.com/tony-42069/council-sim - **Demo video:** https://www.youtube.com/watch?v=3HYCoF0bVFg - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=23 ### 24. Paideia Live URL: https://usepaideia.com/ Paideia unifies planning, grading, IEP compliance, parent communication, and district analytics into one AI-native K-12 platform — built entirely by Claude. Teachers are drowning: 44% report burnout, grading 150 essays takes 37+ hours, SPED teachers spend more time on paperwork than instruction. Parents juggle five apps and can't tell if their kid is learning. Paideia solves this. Grade a class of essays in 90 seconds. Draft a complete IEP in under an hour instead of ten. Parents get plain-language progress updates. Students get a Socratic AI tutor grounded in research showing 2x learning gains. Claude built everything: 10 research subagents explored K-12 in parallel, 5 debated product direction via Agent Teams, Claude Code implemented the full app autonomously in 21 hours (55 endpoints, 40+ pages, 13 AI modules, 192 tests), and Claude Cowork produced the demo video. Zero human-written code. One human with a vision. Claude did the rest. - **Team:** [Mark Nowotarski](https://cerebralvalley.ai/u/markmdev) - **GitHub:** https://github.com/markmdev/paideia - **Demo video:** https://youtu.be/qmGM2yjV0V8 - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=24 ### 25. Juan Olano HiveBoard: Datadog for AI Agents. It is framework-agnostic observability for AI Agents. DEMO LINK: https://hiveboard.net/static/fleet.html?apiKey=hb_live_dev000000000000000000000000000000 Includes a Claude Code Skill for 5-min agent setup and a Claude Agent SDK hook (hiveloop_hooks()) for zero-config observability (shipping imminently) Five views: Fleet (live status, heartbeats, queue depth). Cost (every LLM call by model/agent). Analytics (error clustering, heatmaps, etc). Agent View (per-agent medical chart with full LLM logs), and Insights (38 structured questions with health scoring). Open-source SDK (pip install hiveloop) with 80+ sensors across 3 progressive layers. Claude Code Skill integrates any agentic codebase in ~5 minutes via a slash command. Built in 4 days using 3 Claude Opus 4.6 instances as a coordinated team: PM, and two dev teams cross-auditing each other's code. ~4% coding time; rest was specs, design, and adversarial review. - **Team:** [Juan Olano](https://cerebralvalley.ai/u/jcolano) - **GitHub:** https://github.com/jcolano/hiveboard - **Demo video:** https://youtu.be/lLWr9_1cgNw - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=25 ### 26. VinceOvando LLMitM v2 is an autonomous pentesting agent where Opus 4.6 builds the automation that replaces itself. Penetration testing is locked behind years of expertise and expensive tooling. LLMitM v2 breaks that barrier: it utilizes Anthropic's frontier models to capture HTTP traffic from a target web application and reason about where business assumptions and code behavior diverge — the methodology senior pentesters use intuitively — then compiles that reasoning into a deterministic ActionGraph stored in Neo4j. The first engagement requires Opus 4.6's reasoning. Every subsequent engagement against a matching target executes with zero LLM calls. When a stored graph breaks, the system classifies the failure, Opus 4.6 recompiles only the affected portion, and the fix is stored permanently — reused at zero cost on the next run. Opus 4.6 made this possible: a self-repairing agentic system that bridges natural language interpretation and programmatic automation. - **Team:** [Vince Ovando](https://cerebralvalley.ai/u/vinceovando) - **GitHub:** https://github.com/cybersharkvin/llmitm_v2/ - **Demo video:** https://youtu.be/VNIdEaM3IDI - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=26 ### 27. Art Curator Images are meant to be searched, while art is meant to be explored. This project provides Claude an MCP to browse The Metropolitan Museum of Art's public domain collection. - **Team:** [Edward Gao](https://cerebralvalley.ai/u/edwarddgao) - **GitHub:** https://github.com/edwarddgao/curator - **Demo video:** https://youtu.be/s_pfyLrQWBo - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=27 ### 28. TendHunt TendHunt is an AI procurement intelligence platform where one agent replaces an entire business development team. It ingests 10,000+ live UK government contracts hourly, AI-scores them against your company profile, enriches 2,384 buyer orgs with key personnel and governance data, monitors competitor wins in real-time, and surfaces buying signals from board minutes - before tenders are even published. - **Team:** [Kyrylo Kozak](https://cerebralvalley.ai/u/kozak) - **GitHub:** https://github.com/k1p1l0/tendhunt - **Demo video:** https://youtu.be/llBH3bBFItU - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=28 ### 29. Jon McBee Elisa is a block-based visual IDE where kids design software by snapping blocks together and AI agents build it. Users compose specs using seven primitives; Goals, Requirements, Minions (Agents), Skills, Rules (Hooks), Portals (MCP/CLI/hardware connectors), and Deployments, then hit Go. A meta-planner decomposes the spec into a task DAG, and Builder, Tester, and Reviewer agents execute it with live visibility into planning, coding, testing, and deployment. A teaching engine surfaces age-appropriate programming concepts as agents work. Built solo in one week, entirely with Claude Code. I architected and spec'd; Claude wrote every line. 39,000 lines of TypeScript/Python/React. 1,500 tests. 76 commits. The primary artifacts of software creation are shifting from source code to specs and tests. Elisa makes that future accessible to the next generation. - **Placement:** 2nd Place - **Team:** [Jon McBee](https://cerebralvalley.ai/u/zoid) - **GitHub:** https://github.com/zoidbergclawd/elisa - **Demo video:** https://youtu.be/rsUaz_QAK6o - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=29 ### 30. ChickenDinner RunFestival puts an AI running coach in your earbuds. Pick a persona and go — Claude talks to you in real time throughout your run. A multi-agent system powers every coaching moment. Claude Opus 4.6 serves as Head Coach with four personas: Hype Coach, Calm Guide, Data Nerd, and Storyteller. Two zero-cost rule-based agents — a Pace Strategist and a research-backed Motivation Engine — enrich every prompt without additional API calls. The Motivation Engine implements findings from four peer-reviewed sports psychology papers: six run phases, "how" vs "why" mindset modes, momentum detection across splits, and self-talk cues from endurance research. The voice pipeline achieves <5s latency by streaming Claude's response sentence-by-sentence through OpenAI TTS. Runners can talk back — tap the mic and Claude responds with full context. Community presence shows who's running worldwide right now. Because running alone doesn't mean running lonely. https://runfestival.vercel.app - **Team:** [Spencer Yang](https://cerebralvalley.ai/u/spenceryang) - **GitHub:** https://github.com/spenceryang/RunFestival - **Demo video:** https://www.youtube.com/watch?v=g_RIvPc3VQk - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=30 ### 31. Code Debugger Code Catastrophe Predictor uses Claude Opus 4.6's adaptive thinking to imagine how production systems fail before they do. Unlike traditional testing that only finds what you look for, it generates creative, multi-step cascade failures the kind that cost Knight Capital $440 million in 45 minutes or took down AWS S3 with a single typo. Production failures are creative cascading disasters involving race conditions, load spikes, and edge cases that static analysis and unit tests miss entirely. The Solution: Opus 4.6 analyzes your architecture and generates realistic catastrophe scenarios with step-by-step cascades, impact estimates, and historical precedent from real incidents. A multi-agent system validates each scenario to prevent false positives. I initially built personal projects Telegram bot,automated trading system) during the hackathon but pivoted to test Opus 4.6's creative reasoning, discovering it excels at adversarial production system analysis. - **Team:** [Syed Atif](https://cerebralvalley.ai/u/sydatf) - **GitHub:** https://github.com/syedatif4118/Code-Catastrophe-Predictor/tree/main - **Demo video:** https://youtu.be/hSk4iBB69iQ - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=31 ### 32. Board of Advisors AI Board of Advisors democratizes expert business advice by assembling 8 specialized AI advisors powered by Claude Opus 4.6 that debate, challenge each other, and collaborate in real-time to evaluate your business idea. Describe your concept and watch a CFO, Compliance Officer, Rival Strategist, Customer, Brand Strategist, Operations Chief, Sales Closer, and Product Manager negotiate your plan. Agents ask clarifying questions, form sub-groups, propose agreements, and reach consensus through endorsement voting. Unlike multi-agent systems built for code generation, this is the first use of Agent Teams for social simulation. The result is an actionable business plan with investment breakdowns, risk analysis, timelines, and implementation steps — output that traditionally requires expensive consultants or a board of directors. Built with Next.js 16, Claude Opus 4.6 (extended thinking + web search), real-time SSE, and WhatsApp-style chat UI. - **Team:** [Jonatan Vazquez](https://cerebralvalley.ai/u/jonatan), [Vincenzo Meza](https://cerebralvalley.ai/u/Eijun) - **GitHub:** https://github.com/jonatanvazquez/ai-board-of-advisors - **Demo video:** https://youtu.be/PrpvRGA4Hvg - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=32 ### 33. Spectrix TargetRadar: Drug Target Validation in 30 Seconds Enterprise tools cost $50-200K/year and take weeks. TargetRadar delivers comprehensive analysis—free, instant, open source. FEATURES: • 7-Dimension Scoring: Genetic Evidence, Chemical Tractability, Structural Readiness, Clinical History, Regulatory Genomics (AlphaGenome), Literature, Innovation—querying Open Targets, ChEMBL, PDB, ClinicalTrials.gov, Ensembl, PubMed, bioRxiv • 16 AI Features (Opus 4.6): Intelligence briefs, competitive analysis, hypothesis generation, conflict detection, trial design • Tools: Molecule Testing Lab, DeepTarget Discovery, 3D visualization, docking, .docx/.xlsx reports Live: http://138.68.7.97:4000 MIT licensed, no login required. - **Team:** [Heinz Jungbluth](https://cerebralvalley.ai/u/jgheinz) - **GitHub:** https://github.com/Bionicsdemo/targetradar - **Demo video:** https://youtu.be/e2dImTZ4Q9Q - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=33 ### 34. Eddie Liu People spend 5–6 hours a day on their phones (30–40% of their waking life) often mindlessly scrolling. Existing screen time apps focus on reduction, but never answer: to what end? Lucid is an iOS app that helps users redirect mindless screen time into activities aligned with their values. During onboarding, you describe your typical day by voice. Claude analyzes it and reveals how you actually spend your time. Then you set a phone usage goal and describe your ideal life. The AI generates a personalized daily template tracking the things that matter to you. Each morning, Claude helps you plan your day. Throughout the day, a DeviceActivity Monitor tracks your mindless apps, and a Circuit Breaker shield fires when you exceed your goal, reminding you of your own intentions. Each evening, Claude provides CBT-informed coaching based on the gap between intention and reality. Over time, the gap shrinks. Built entirely with Claude Code in one week! - **Team:** [Eddie Liu](https://cerebralvalley.ai/u/eddieliu) - **GitHub:** https://github.com/eddielement/Lucid - **Demo video:** https://youtu.be/el0ElvmWKvk - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=34 ### 35. Astro MatchPoint is a two-sided marketplace connecting businesses (Seekers) with startups (Builders) to solve real challenges, orchestrated by a multi-agent AI system powered by Claude. Seekers use a chat-based Discovery agent (Sonnet 4.5) to uncover pain points through natural conversation. Opus 4.6 then synthesizes discussions into concise product specs. A Scout agent (Haiku 4.5) searches 5,690+ startups, delivering ranked matches with AI explanations. Users can save, compare, and connect via magic-link invites. Builders gain visibility into aggregated market needs and trends, with an AI-guided workspace powered by Opus 4.6 and E2B sandboxes for MVP development. The platform runs on a unified state machine with real-time SSE streaming and full observability. - **Team:** [Arthur Granja](https://cerebralvalley.ai/u/turo) - **GitHub:** https://github.com/artcgranja/MatchPoint - **Demo video:** https://youtu.be/J0qPBJVuYDw - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=35 ### 36. ChainReview Vibe coding makes it ridiculously easy to ship. You generate features in minutes, hit run, and everything looks fine—until the first real edge case, the first PR review, the first CI failure. Then the "almost right" code shows up as noise: false positives, shallow findings, and hours lost in back-and-forth. We built ChainReview to turn that chaos into a closed loop. ChainReview is a multi-agent AI code reviewer where five specialists—Architecture, Security, Bugs, Explainer, and a strict Validator—work in parallel. Every agent is grounded in deterministic tooling via CRP (ChainReview Protocol), an open MCP-compatible schema with 20+ tools: file reads, import graphs, call graphs, Semgrep scans, and ripgrep pattern matching. Every finding is pinned to real evidence—file paths, line numbers, and confidence scores. The twist is the Validator Agent. It challenges every finding, re-reads the code, checks context, and kills the noise. What survives is real, actionable, and ready to fix. - **Team:** [Charann Srinivas](https://cerebralvalley.ai/u/charann) - **GitHub:** https://github.com/charannyk06/ChainReview - **Demo video:** https://youtu.be/MVYIiM0yKTc?si=dfm4F4klorZ4F_Fi - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=36 ### 37. Quincy Labs TRACE is actionable preventative causal intelligence for chronic disease. Instead of telling patients what they already know — that they're in pain — TRACE traces why, connecting symptoms across body systems to their biological root causes. I built this because I have sickle cell disease. A sinoplasty fixed my chronic back pain. No doctor connected sinus inflammation to sickling pain in my spine. TRACE is the tool I needed — one that reasons across systems the way the disease actually moves. Three Claude agents power a real-time pipeline. Sonnet 4.5 parses natural language symptoms. Opus 4.6 with extended thinking traces causal chains across pathophysiology, autonomously pulling evidence through Exa and FireCrawl MCP tools. A second Sonnet 4.5 agent generates urgency-rated, actionable recommendations. - **Team:** [Henry Ogedegbe Jr](https://cerebralvalley.ai/u/danchou) - **GitHub:** https://github.com/zetsuchan/Trace - **Demo video:** https://youtu.be/htvT8ynIe6A - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=37 ### 38. Timothy Nice Popcorn is a Chrome extension and Claude Code integration that accelerates verification of AI-generated UI changes while reducing token usage and latency. Instead of relying on many small LLM calls to drive the browser, Popcorn batches workflows and executes them natively inside Chrome, returning structured evidence such as assertions, screenshots, and expected versus actual results to Claude in bulk. When Opus edits frontend code, Popcorn automatically generates verification flows based on plain-English acceptance criteria. Bugs surface immediately, Claude can self-correct, and engineers can observe, review, and replay recorded runs. Popcorn keeps humans in the loop while making AI-driven development faster, cheaper, and more trustworthy. - **Team:** [Timothy Nice](https://cerebralvalley.ai/u/ser3ph) - **GitHub:** https://github.com/timothynice/popcorn - **Demo video:** https://youtu.be/iNs_Dlb2WKY - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=38 ### 39. Prism Prism is the first AI-native system for product discovery. While coding agents like Cursor and Claude Code solve how to build software, Prism solves the harder problem: deciding what to build. Product discovery is often disconnected and buried in spreadsheets. Prism changes that. It ingests customer interviews, usage data, and tickets to answer "What should we build next?" with rigorous, data-backed simulations. Prism proactively identifies trends and runs agentic market simulations (e.g., "What if we shift to usage-based pricing?"), predicting revenue impact with confidence intervals. It explains the "why" behind decisions using transparent reasoning chains and closes the loop by generating implementation-ready artifacts, PRDs, Linear tickets, and specs, for your coding agents. Prism turns product strategy into code in seconds. - **Team:** [Krish Garg](https://cerebralvalley.ai/u/krishhhg) - **GitHub:** https://github.com/kgarg2468/Cursor-For-PMs - **Demo video:** https://youtu.be/mHWURXnNmyo - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=39 ### 40. Tacit Tacit extracts unwritten rules your team enforces but never documented. Point it at a GitHub repo — 16 Agent SDK agents mine PR reviews, CI failures, and CHANGES_REQUESTED discussions, then output CLAUDE.md and `.claude/rules/` files Claude Code loads automatically. The key: anti-pattern mining from rejected PRs extracts "NEVER do X" rules linking each to the exact PR comment where the team learned the lesson. 57 agent runs, 20 MCP tools, ~$13/extraction. 4-phase pipeline with Opus for deep analysis, Sonnet for scanning. Results: 343 rules from 50 PRs, 100% with provenance. Eval suite: 83%. Rediscovered 87% of documented guidelines from PR data alone — without reading the existing CLAUDE.md. - **Team:** [Bayram Annakov](https://cerebralvalley.ai/u/Bayka) - **GitHub:** https://github.com/BayramAnnakov/tacit - **Demo video:** https://youtu.be/JEyUfCVTNN4 - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=40 ### 41. AJ Green Every AI web agent takes a screenshot, ships it to a vision model, waits 5-10 seconds, burns $0.05, and gets back a guess. Every click, every scroll, the same expensive, slow loop. Ghost reads the DOM and generates structured MCP tools in under 2ms. 1,250x faster than screenshot agents. Extractors, navigators, form handlers, all cached locally. First visit builds the tools. Every visit after is instant and free. When sites change, Ghost heals itself. 188 domains. 1,399 tools cached and growing. Ghost gets smarter every time you use it. Fully local, zero marginal cost, zero cloud dependency. Every other agent pays per action. Ghost pays once and compounds forever. - **Team:** [AJ Green](https://cerebralvalley.ai/u/AJSAI) - **GitHub:** https://github.com/ajsai47/ghost - **Demo video:** https://youtu.be/5GNMhwikZp4 - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=41 ### 42. SimForge Building simulation environments for robotics is painfully manual. SimForge converts a single photo into a physics-ready robotics simulation using 6 AI agents orchestrated by Claude. Each agent handles a stage of the pipeline, from object detection and dimension estimation to MuJoCo XML generation, physics validation, and visual verification through an iterative feedback loop. The result is a sim2real-ready training environment built from a snapshot in a few minutes. - **Team:** [Jake Kang](https://cerebralvalley.ai/u/kangjake) - **GitHub:** https://github.com/jacquekang/claude-code-hackathon - **Demo video:** https://drive.google.com/file/d/1hEETnaVDm9BZCP2bLMLQVQixRq3wjJXy/view?usp=sharing - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=42 ### 43. Team Hey Jamie Podcasting is hard. It's hard enough for me just to talk extemporaneously about current events or topics without also having to juggle browsing the web to fact check or demonstrate something related to the discussion at the same time. Many of the top podcasts have producers that handle this part of the production. Not all podcasters are lucky enough to afford a producer or have a friend willing to record a podcast with them at any hour of the day. Not any more! HeyJamie is an autonomous podcast producer that helps any podcaster have a killer show. HeyJamie can also take on different personas and act as a Fact Checker, a Debunker or even a Comedian. All of this is controlled by a prompt given to Claude who acts as the planner throughout the show. My hope for this project is that podcasters will be able to up their podcasting game through the use of this tool and we can all benefit from a podcast producing agent for everyone! - **Team:** [David Norman](https://cerebralvalley.ai/u/hypercatcher) - **GitHub:** https://github.com/normand1/HeyJamie - **Demo video:** https://youtu.be/3mxlBCtPjf4 - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=43 ### 44. Beacon When someone you love is diagnosed with a rare disease, you're handed a name you can't pronounce and told there's no cure. Beacon replaces that lonely fight with a full AI team that activates instantly. Nine Opus 4.6-powered agents span the entire pipeline from basic science to clinical action: mapping disease mechanisms, screening compounds for drug repurposing, evaluating safety, drafting researcher outreach, mapping FDA pathways, identifying funding, and synthesizing it all into a treatment roadmap. The family member stays in control, but thousands of hours of research are handled. No family should have to fight alone. Judges can access hosted app with provisioned API key from: http://beacondemo.vercel.app/?token=beacon2026 Everyone else can access from https://beacondemo.vercel.app/ (have to provide Anthropic API key) Demo mode (fast) can be seen at https://beacondemo.vercel.app/?demo - **Team:** [Shawn Dimantha](https://cerebralvalley.ai/u/shawnd) - **GitHub:** https://github.com/shawndimantha/beacon - **Demo video:** https://www.youtube.com/watch?v=Co0OdXqUr-g - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=44 ### 45. ButterFence ButterFence is a Claude Code-native safety harness that red-teams your repo against destructive and secret-leaking AI agent behavior, then auto-generates enforceable hooks to prevent it. The core innovation: Claude Opus 4.6 as the adversary. The redteam command scans your repo structure and tech stack, then asks Opus 4.6 to generate creative attack scenarios, obfuscation, encoding tricks, variable indirection, tailored to YOUR codebase. ButterFence catches them in real-time. When attacks get through, Opus 4.6 analyzes gaps and generates exact regex patches. One command runs the full attack-patch-verify loop. Ships with 44 scenarios, 11 threat categories, 7 rule packs (OWASP, AWS, Docker), 5 export formats, entropy detection, obfuscation decoding, behavioral chain detection, live dashboard, and CI/CD integration. Claude attacks. Claude defends. Claude patches. 315 tests. 35+ modules. Open source. - **Team:** [AYUSHMAN Mukherjee](https://cerebralvalley.ai/u/Ayushman) - **GitHub:** https://github.com/ayush585/ButterFence - **Demo video:** https://www.youtube.com/watch?v=WM6_JolSBSA - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=45 ### 46. Vaidya AI There is a huge gap in the patient to doctor ratio in India and especially in the rural villages, patients have no access to even primary care. LLMs like Claude are great at reasoning but are not accessible to remote villages due to language barriers. In addition to this doctors don't have access to the latest agentic workflows to update EHRs and get SOAP notes. Second opinion AI is a completely agentic app that is multi-lingual uses routing to optimize for extended thinking and automatically compresses context from huge EHRs Doctor Portal: https://second-opinion-ai-doctor.vercel.app Login: Email: doctor@demo.secondopinion.ai password: doctor123456 Patient portal: https://second-opinion-ai-patient.vercel.app Login: Email: demo@secondopinion.ai Password: demo1234 - **Team:** [Nikhil Nakhate](https://cerebralvalley.ai/u/nikx-vla) - **GitHub:** https://github.com/nikhil-nakhate/second-opinion-ai - **Demo video:** https://www.youtube.com/watch?v=48jeKhrH_ss - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=46 ### 47. Long Nights Agent Harness Builder is a visual tool for orchestrating multi-agent AI workflows. It solves the "this" problem: when people say "build this," frontier models handle the build, but defining and structuring the "this" is where things break down. Long-running tasks lose context. AI drifts. Inspired by Anthropic's research on long-running task harnesses, the goal is to bring structured AI orchestration to anyone with expertise, not just devs. You create projects with structured tasks and step-by-step checklists that keep agents anchored. Three AI abstractions layers: Skill Generation captures domain expertise, Process Team designs collaboration workflows, and Task Initialization breaks down projects. Then multi-agent orchestration executes it all with real-time tracking, cost monitoring, and AI-generated summaries Works across domains. You bring your expertise. The harness structures it. AI handles the build. Note: Git commits under jarvis-sam CC agent account - **Team:** [Sebastian Almeida](https://cerebralvalley.ai/u/sebasalmeidam) - **GitHub:** https://github.com/sebasalmeidam/agent-harness-builder - **Demo video:** https://www.loom.com/share/05d997586e154c9a83d51badddf9f777 - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=47 ### 48. micsparre xray reveals who really owns your code... and where that's a problem. Engineering teams make staffing and risk decisions on gut feel because the real knowledge topology is invisible. Commit counts can't tell you if someone is an architect or just fixing typos. When a key engineer leaves, you find out too late that entire systems depended on them. xray uses Claude to read actual PR diffs, classify expertise depth per contributor per module, evaluate review quality, and use extended thinking to surface hidden patterns — ghost architects, knowledge silos, review rings. A single bus factor blindspot can cost months of productivity. xray makes that risk visible and actionable before it becomes a crisis. - **Team:** [Michael Sparre](https://cerebralvalley.ai/u/michaelsparre) - **GitHub:** https://github.com/micsparre/xray - **Demo video:** https://www.youtube.com/watch?v=ddSjYo4Gc2A - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=48 ### 49. botsec Forge Proof combines Claude Opus with formal verification to find smart contract vulnerabilities with mathematical proof. One command runs a two-stage agentic pipeline on the Claude Agent SDK with specialized sub-agents, constrained tools per agent, structured JSON output, and Skills for preloading FV domain knowledge: Threat Model — An Opus agent explores the codebase using pre-computed static analysis (call graphs, CEI ordering, auth checks, AST), identifies threats from attacker perspectives, and outputs a ranked threat model with code traces. Analyze — A second Opus agent writes formal verification tests, runs SMT solvers across all possible inputs, interprets counterexamples to distinguish real bugs from bad specs, and falls back to fuzz testing when the solver hits bounds. Tested against 3 production DeFi protocols with active bug bounties: 41 formally verified proofs and 4 violations with concrete counterexamples. Total cost: ~$47. Vulnerabilities were responsibly disclosed - **Team:** [Anto Joseph](https://cerebralvalley.ai/u/blocksec) - **GitHub:** https://github.com/antojoseph/botsec - **Demo video:** https://www.youtube.com/watch?v=5kvLsW2udVg - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=49 ### 50. Nexus Nexus is an autonomous AI engineering team that turns one instruction into verified code patches, eliminating coordination busywork between PMs, frontend, and backend engineers. Six Claude Opus 4.6 agents work together: Router decides focus, Frontend and Backend agents scan with domain expertise, Manager prioritizes like a PM, Detect reviews code deeply, Plan creates tasks, and Patch generates real unified diffs applied via git and verified with a real build. Opus 4.6 powers six distinct steps with different system prompts, extended thinking visible in the UI, and structured JSON contracts between agents. A Reflector saves learnings to persistent memory so every run is smarter, and the system also learns from external articles automatically. No hardcoded rules, every decision is Claude. The demo shows three flows on a real Next.js codebase: scan mode, instruction mode with auto-apply and build verification, and continuous learning from the web. - **Team:** [Rashmi E](https://cerebralvalley.ai/u/rashmi), [Gokul Srinath Seetha Ram](https://cerebralvalley.ai/u/gokul) - **GitHub:** https://github.com/rashmi-star/claude-hackathon - **Demo video:** https://www.youtube.com/watch?v=KefaOsdAxKw - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=50 ### 51. Tolani Omitokun The Emergent Diagnostic Institution is a six-agent AI system built with Claude Opus 4.6 and Claude Code. Three specialists independently analyze complete medical histories, then engage in multi-round debate. A Metacognitive Observer detects cognitive biases like anchoring and premature closure. A Patient Translator writes plain-language explanations for families. A Constitution Amender rewrites the system's rules after every case. It processes 77,000+ words of medical records (not summaries), agents challenge each other through structured disagreement, and the system evolves after every case, generating 43 amendments across 6 evaluations. The Observer caught 24 cognitive biases. In the showcase case, EDI analyzes a boy misdiagnosed with autism for 12 years across 8 doctors and 3 states, identifying CLN2 Batten Disease with 91% confidence. 100% accuracy across 6 domains. Built in 3 days, solo. Live: https://emergentdiagnostic-ai.up.railway.app (Code: DEMO2026) - **Team:** [Tolani Omitokun](https://cerebralvalley.ai/u/tolatokuns) - **GitHub:** https://github.com/tolaniomitokun/emergent-diagnostic-institution - **Demo video:** https://youtu.be/sLp82UfXZXk?si=dMBEb2YUcB6nlDC4 - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=51 ### 52. Build but Don't Code Capture is available to public at https://capture-app.xyz Capture is an AI-native note-taking app designed around one idea: save anything and everything whenever you want and let Capture figure it out. Whether it's a link you stumbled on, a thought you want to hold onto, a PDF you need to revisit, a screenshot worth remembering, or something you'd rather just say out loud — Capture stores it, summarizes it, categorizes it, and then nudges you toward what to explore next. 1. The idea is to remove friction from note taking. And more importantly, make it less sparse and scattered. 2. With auto AI summarization and relevant recommendation generation, this will allow anyone to keep context of the notes they make not only then but forever. 3. With various media support, you can write your thoughts, upload pdfs and images, and drop any number of links and let the AI engine organize them for you. 4. Certainly, in this day and age, it comes with Voice as a primary medium of interaction. - **Team:** [Gaurav Jain](https://cerebralvalley.ai/u/gauravjain14) - **GitHub:** https://github.com/gauravjain14/capture-claude - **Demo video:** https://youtu.be/dWluSDILjn8 - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=52 ### 53. Nour Desouki FenceFlow eliminates paper-based workflows from fencing tournaments using Opus 4.6. Built from hosting 30+ events, it solves a real problem: staff squint at illegible handwriting, results lag hours, and errors change seedings. Four interfaces: The Committee Dashboard runs an autonomous agent that auto-approves scoresheets, flags anomalies for human review, and logs decisions with reasoning. The Referee Portal uses Vision OCR with extended thinking to extract scores from photographed handwritten sheets. The Coach View runs a Bradley-Terry model with Monte Carlo simulations and multi-turn Opus chat for advice. The Public View streams live brackets and AI narrator commentary token by token. Fencing software lacks public APIs so the demo uses data from a tournament the developer hosted. - **Placement:** Finalist - **Team:** [Nour Desouki](https://cerebralvalley.ai/u/ndesouki) - **GitHub:** https://github.com/nourdesoukizz/Cozmx-fencing- - **Demo video:** https://youtu.be/Yai0l43kM9c - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=53 ### 54. Mike CrossBeam tackles California's ADU permit crisis. 429,000+ permits issued since 2018, 90%+ sent back for corrections — each cycle costing weeks and thousands of dollars. Upload your corrections letter and construction plans. CrossBeam reads blueprints using vision, interprets every correction with code citations, cross-references state and city law via live web research, and drafts a response package ready for your engineer to sign. Built on the Claude Agents SDK with 13 custom skills covering state ADU law, city-level research, blueprint extraction, and corrections interpretation — all running concurrently. - **Placement:** 1st Place - **Team:** [Mike Brown](https://cerebralvalley.ai/u/brz) - **GitHub:** https://github.com/mikeOnBreeze/cc-crossbeam - **Demo video:** https://youtu.be/jHwBkFSvyk0 - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=54 ### 55. Gatz Most AI dev tools try to own the whole pipeline — prompt, generate, deploy, done. I don't buy it. When your project gets real, all-in-one tools fall apart. Baaton takes the opposite bet: your agent is already good at coding. What's missing is the layer above. An orchestration board where AI agents (Claude Code, Codex) are first-class. They connect via REST API, pull tickets, update status, and post structured summaries — files changed, test results, PR links. The human stays in control. - Kanban with drag-and-drop, bulk actions, shortcuts - Agent API with key-based auth - TLDR system — agents post structured debriefs after work - Public forms with LLM auto-qualification for bugs - Multi-tenant via Clerk orgs with Postgres RLS - Cross-org dashboard across all your teams. Stack: React 19, Vite 8, Rust/Axum, Supabase Postgres, Clerk, Railway. ~30 endpoints with compile-time checked SQL. The future isn't one magic tool. It's specialized agents on a shared board. Baaton is that board. - **Team:** [Ramzi Lb](https://cerebralvalley.ai/u/rmzlb) - **GitHub:** https://github.com/rmzlb/baaton - **Demo video:** https://youtu.be/usX1k144okA - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=55 ### 56. TARA Africa needs billions annually in transport infrastructure but most projects fail at feasibility. Road appraisal takes weeks of specialist work and most agencies lack capacity. In Uganda we spend over $1B yearly on roads with $20M on planning and the process is manual and inefficient. TARA transforms this by allowing us to use dashcam footage. Claude Opus 4.6 analyses every frame to give surface condition and roadside activity. It segments the road, selects interventions with Uganda-calibrated costs, and runs full cost-benefit analysis with NPV, EIRR, BCR, sensitivity and deterioration modelling. What makes TARA different: it asks who benefits. The equity module identifies pedestrians without footpaths, school children on the carriageway. It shifts appraisal from economics to human-centred analysis. Built with Claude Opus 4.6 for vision, condition narratives, sensitivity and equity. Tested on real dashcam footage near Kampala. Open source. - **Placement:** Finalist - **Team:** [Kyeyune Kazibwe](https://cerebralvalley.ai/u/KyeyuneKazibwe) - **GitHub:** https://github.com/Kye256/tara-transport-assessment - **Demo video:** https://www.youtube.com/watch?v=GFCrXehS1DE - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=56 ### 57. Jiggle Wiggle Jiggle Wiggle is a real-time AI movement coach that works with three input modes: YouTube videos, live Zoom calls, and AI-generated references. Paste any dance or fitness video URL and the app downloads it, extracts reference poses frame-by-frame, and compares your webcam feed against the dancer in real-time with split-screen skeleton overlays and per-body-part scoring (arms, legs, torso). In Zoom mode, it captures an instructor's video feed and runs pose detection on both streams simultaneously, coaching you through matching their moves. An LLM-powered voice coach provides adaptive spoken feedback in real-time with specific limb-level corrections. Hand gesture controls let you interact without touching the keyboard. All pose detection runs entirely in-browser via MediaPipe WASM with no server-side GPU needed for the core experience, while SAM2 runs on Modal's A100 GPUs for video segmentation. Built with Next.js 16, React 19, TypeScript, OpenAI GPT-4o-mini, Groq, Zoom Meeting SDK. - **Team:** [Aryan Keluskar](https://cerebralvalley.ai/u/aryankeluskar) - **GitHub:** https://github.com/cindyzli/jigglewiggle - **Demo video:** https://www.youtube.com/watch?v=kL3udtWY1EE - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=57 ### 58. StringAI Cowpy is a cat that turns any live website into a production-ready Next.js 15 app from a single URL. Unlike screenshot-to-code tools that ask an LLM to guess markup from a static image, Cowpy runs an 8-step agentic pipeline: it crawls the sitemap, records browser sessions as video, captures HAR network traffic, detects animations frame-by-frame via video analysis, and generates a full Next.js app The result is a real, buildable app with routes, animations, hover states, and API integrations — not a static HTML approximation. The entire pipeline is orchestrated through Claude Code skills and subagents, coordinating Claude Opus 4.6, Sonnet 4.5, Haiku 4.5, Kimi K2.5 and Flux across specialized roles. - **Team:** [Dias Khalniyasov](https://cerebralvalley.ai/u/dias-kh) - **GitHub:** https://github.com/DiasKhalniyasov/full-clone-site2code - **Demo video:** https://www.youtube.com/watch?v=j71GZ7MUp2c - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=58 ### 59. Team Statistics I built PredictFlow because I kept staring at Polymarket thinking that price looks wrong, but I had no way to prove it or decide how much to risk. PredictFlow pulls live data from Polymarket and Kalshi, runs markets through a three model ensemble, calculates the edge after fees, and automatically paper trades with guardrails. Users stay in control without babysitting. I trained models on 4,341 resolved markets using a strict temporal split. While building I found leakage in volume features with a 0.85 correlation to outcomes. I learned and removed them, and documented it. The cleaned system still beats baseline by 21.5%, especially in the 20 to 80 percent range where trades happen. Claude Opus 4.6 returns structured probability, drivers, and risks. Requests are SHA-256 cached, so repeats cost zero. Under the hood are PostgreSQL, many APIs, and a loop that collects, predicts, trades, and scores itself. I started with a signal page. I went above and beyond and built a small quant desk. - **Team:** [mian abdullah](https://cerebralvalley.ai/u/mian) - **GitHub:** https://github.com/mian-abd/prediction-market-analysis - **Demo video:** https://youtu.be/W_PEd88-Y10 - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=59 ### 60. Henk van Ess (solo) Gym memberships, app terms, pet adoptions, sweepstakes — the things you sign with a shrug. They weren't written with one. FlipSide treats a $2 coupon with the same seriousness as a $500,000 corporate lease. Existing legal tech targets lawyers. Try finding one that handles gym memberships or sweepstakes. Upload any document or pick from 14 samples. Flip cards: the front shows what the drafter wants you to feel. Flip it to see what an expert catches. Risk scores, trick types, dollar figures. Why Opus 4.6? Haiku builds cards fast and parallel. The verdict needs Opus: cross-clause reasoning connecting a liability waiver on page 2 to an arbitration clause on page 8, and theory of mind that reconstructs drafter intent. Millions sign one-sided documents without seeing the other side. Now they can. 0% human code. 14,000+ lines built through conversation with Claude Code. 116 commits. 6 days. - **Team:** [Henk van ess](https://cerebralvalley.ai/u/henkvaness) - **GitHub:** https://github.com/voelspriet/flipside - **Demo video:** https://vimeo.com/1165331217 - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=60 ### 61. Jarrod Watts Scuttle Pay is the bank account for AI agents; it lets AI agents buy things online - no humans (or credit cards) required. - **Team:** [Jarrod Watts](https://cerebralvalley.ai/u/jarrodwatts) - **GitHub:** https://github.com/jarrodwatts/scuttlepay - **Demo video:** https://youtu.be/OKR_YtCgcXo - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=61 ### 62. zzappa Gradient is a language learning Claude-powered app that transforms English text progressively into a target language through 7 levels of "gradient immersion". Users read a story that gradually shifts from English to target language — grammar first, then vocabulary — so the target language's structure should become intuitive before new words appear. The reader interface features inline annotations with translations, pronunciation (IPA), and grammar notes in a synchronized side panel. Terms are explained only on first occurrence across the entire project. Additional tools include side-by-side comparison mode, contextual grammar chat (double-click any word), comprehension quizzes, text-to-speech, and spaced-repetition flashcards — all generated automatically from the reading content. Claude can answer any questions on grammar and vocabulary, and prepare personalized study sessions and quizzes. - **Team:** [Artem Golenchenko](https://cerebralvalley.ai/u/zzappa) - **GitHub:** https://github.com/zzappa/gradient_reading - **Demo video:** https://youtu.be/53Ojkimo5fE - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=62 ### 63. Shipwell Shipwell is a full codebase autopilot that ingests entire repositories into Opus 4.6's 1M-token context window for deep cross-file analysis. It runs five operations - security audit, migration planning, refactor analysis, documentation generation, and dependency upgrades, also streaming findings in real time with severity tagging, cross-file dependency graphs, and before/after code diffs. One click creates a fix PR through our GitHub App with all patches applied. Available as both a web dashboard (shipwell.app) and a CLI (npm i -g @shipwellapp/cli). Unlike single-file linters, Shipwell sees your entire codebase at once, catching issues that span multiple files, like auth bypasses across route handlers or data flows through three modules. - **Team:** [MANAS DUTTA](https://cerebralvalley.ai/u/manasdutta) - **GitHub:** https://github.com/ShipwellHQ/shipwell - **Demo video:** https://youtu.be/3xRUQDbKKSA - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=63 ### 64. iy3r SEBI's CSCRF mandates cybersecurity compliance for every regulated entity in India's securities market. The framework spans seven functions across 200+ pages of standards and guidelines. Compliance typically requires expensive consultants and months of manual work. I built a complete compliance-as-code platform using Claude Code. The entire CSCRF is extracted verbatim into structured Markdown with YAML frontmatter—every standard, guideline, and applicability rule preserved exactly. 3 Claude Code skills (/ciso, /ciso-policy, /ciso-assess) orchestrate five specialist agents delivering a full compliance lifecycle: entity profiling, gap analysis, policy generation with regulatory traceability, and a 90-day remediation roadmap. The system is context-efficient, quality-gated, and deterministic. A validation suite ensures correctness. What once took months and tens of thousands of dollars now takes a single conversation. - **Team:** [Vashistha Iyer](https://cerebralvalley.ai/u/iy3r) - **GitHub:** https://github.com/iy3r/sebi-cscrf - **Demo video:** https://youtu.be/_bBc9DCljPI - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=64 ### 65. Sahaib Sehat: voice-first multilingual medical triage for India's 700M underserved (PS2: Break the Barriers). Speak symptoms in 7 Indian languages, Opus 4.6 extended thinking performs visible clinical reasoning, classifies severity, generates bilingual action plans with printable doctor cards. Opus 4.6 does what no other model could: structured medical reasoning across code-switched languages (Hinglish, Tanglish), WORST-FIRST triage mirroring clinical ESI/MTS, culturally aware warnings on dangerous home remedies. The thinking chain IS the product. Two-layer emergency detection catches emergencies in <50ms. Period Health Companion breaks a deeper barrier: 71% of rural Indian girls don't know about menstruation before their first period. Women log cycles, get AI predictions, ask about PCOS in their language. First explanation without shame. Iterated from text triage to voice loops, emergency bypass, period health, dashboards. Solo build with Claude Code + Sarvam AI for Indian STT/TTS. - **Team:** [Sahaib Singh](https://cerebralvalley.ai/u/sahaib) - **GitHub:** https://github.com/sahaib/sehat - **Demo video:** https://youtu.be/TyzqRgtLUBY - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=65 ### 66. Qdee Intent Guard is an AI secured intent liquidity market that makes safe crypto trading accessible to everyone,no expertise required. DeFi users face hidden dangers like MEV sandwich attacks, toxic slippage, and unfair pricing. Detecting these usually demands deep technical knowledge. With Intent Guard, traders simply state what they want to swap and choose a protection level (Low/Normal/High). Behind the scenes, three solver agents compete to deliver the best execution, while Claude Opus 4.6 analyzes every quote for manipulation, anomalies, and unsafe slippage before funds move. Our hybrid safety model blends hard constraints (min output, gas bounds, slippage limits) with AI reasoning to catch threats rules alone miss. Risky quotes are automatically disqualified. Built end to end with Claude Code: ERC20 escrow contracts, multi-source pricing API, and a wallet-connected swap UI live on Base Sepolia with on-chain proof. - **Team:** [Abdulqodir Iyanda](https://cerebralvalley.ai/u/qdee) - **GitHub:** https://github.com/dolepee/ilm-intent-router - **Demo video:** https://www.loom.com/share/51be89e96104489c98d8921b042a9936 - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=66 ### 67. ZenML ZenML's production website — 2,224 pages across 20 CMS collections — was stuck on Webflow. No version control, no CI/CD, expensive, and migrating manually would take months. I built the replacement during this hackathon using Claude Code with Opus 4.6. It runs on Astro v5 + Cloudflare Pages: content exported via Webflow API, transformed to Markdown, validated against 20 Zod schemas, deployed as ~2,224 static pages in 33s. Seven Preact islands handle interactivity, including a faceted research hub with Pagefind search across 1,453 entries. Live at zenml-io-v2.pages.dev. Opus handled the hard parts: it audited its own schema plan and caught 7 issues before I'd built a page. It noticed Webflow's HTML would break MDX parsing and switched all 1,904 files to standard Markdown unprompted. After hitting parity with the original it went further — redesigning the LLMOps database with features Webflow couldn't support, then fixing its own accessibility bugs. 120 commits, 6 phases, open source. - **Team:** [Alex Strick van Linschoten](https://cerebralvalley.ai/u/strickvl) - **GitHub:** https://github.com/zenml-io/zenml-io-v2 - **Demo video:** https://www.youtube.com/watch?v=vd70OvKFbVc - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=67 ### 68. Capiis Capiis is a local-first wealth management “agent swarm”. Who: HNW individuals (and advisors) handling complex portfolios, taxes, and account/legal structures. What: A read-only dashboard + background agents that generate tax briefings, portfolio insights, price updates, and intel signals. Why: CLI output can be detailed, but without a visual framework users don’t know what to ask next across assets/liabilities/tax. Capiis makes the system legible while keeping humans in control. How: Vanilla JS dashboard (no build step) + Node/Express REST API + SSE realtime refresh. Local data layer = Excel + JSON files. Agents: Tax Analyst, Price Tracker, Feed Analyst. Skills: Onboarding, Tax Professional, Portfolio Intelligence. Tax figures are fact-checked via Perplexity MCP. Built from scratch during the hackathon. More details: DEVLOG.md + CLAUDE.md https://github.com/hanniehoney/Capiis/blob/master/DEVLOG.md - **Team:** [Hannie Liu](https://cerebralvalley.ai/u/hannie) - **GitHub:** https://github.com/hanniehoney/Capiis - **Demo video:** https://youtu.be/8f0CH2Tf1Dg - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=68 ### 69. Silene Systems Website: https://silene.systems Dataset created during event: https://zenodo.org/records/18662232 There are over 7,000 rare diseases affecting 300+ million people worldwide. Their rarity makes them hard to diagnose, difficult to study, and nearly impossible to build clinical tools for. The evidence base is fragmented across thousands of individual case reports in medical literature. Silene Systems automates the extraction of structured clinical data from published case reports using a multi-agent AI pipeline with Claude Opus 4.6, and presents it through an AI-native research workbench, where researchers can browse patients, analyze cohorts, search for digital twins, add their own patients and draft case studies. - **Team:** [Adrian Michalski](https://cerebralvalley.ai/u/AMichalski) - **GitHub:** https://github.com/amichalski2/silene-case-study - **Demo video:** https://www.youtube.com/watch?v=NiARJzu_dYs - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=69 ### 70. TechDebtAssassin TechDebtAssassin is an autonomous AI agent that eliminates the "janitorial" busywork of technical debt. Unlike standard coding assistants that wait for prompts, TechDebtAssassin is a proactive specialist that modernizes entire repositories without human intervention. Powered by Claude Opus 4.6 and the Model Context Protocol (MCP), the agent builds a Global Project Context to reason across file boundaries—inferring complex type hints and architectural dependencies that local-context LLMs miss. It features a robust "Self-Healing" pipeline: scanning the AST for debt, generating type-safe patches, applying them physically to the disk, and running a final verification loop (using mypy) to ensure 100% code health. Built for Problem Statement One, TechDebtAssassin transforms hours of manual refactoring into a one-click autonomous process, turning codebase maintenance from a chore into a reliable DevOps stage. - **Team:** [chiranjeevi joshi](https://cerebralvalley.ai/u/Chiranjeevijoshi) - **GitHub:** https://github.com/jchiru21/tech-debt-assassin - **Demo video:** https://www.loom.com/share/f4ad1749a4e94fc48f70bd23413daead - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=70 ### 71. Brendan from Boston Every year, 1.5 billion people cross international borders, yet 1 in 4 visa applications is refused due to preventable errors like incomplete forms or contradictions. Across 38,000+ travel corridors, each with unique rules, navigating the process is unnecessarily hard. Traverse addresses this with three coordinated AI agents running on Claude Opus 4.6 with extended thinking. The Research Agent searches live government sources for specific requirements. The Document Intelligence Agent reads documents in 40+ languages to find gaps. The Advisory Agent synthesizes this into prioritized fixes. A TypeScript orchestrator coordinates these agents, streaming real-time results via Server-Sent Events and dynamically allocating reasoning depth. Built with Claude Code, Traverse showcases web search, vision, and multi-agent coordination to solve a problem affecting billions of travelers. By mimicking an experienced immigration team, it provides a solution with AI that does not currently exist. - **Team:** [Brendan Erhard](https://cerebralvalley.ai/u/bme3412) - **GitHub:** https://github.com/bme3412/traverse - **Demo video:** https://www.youtube.com/watch?v=qVGSaIUvm5A - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=71 ### 72. Signal-to-Noise Denoise It is an AI engine that decomposes any content (tweet, headline, article, claim, or URL) into its objective truth vectors (verified facts, statistics, and attributions scored against primary sources) and its affective noise (emotional language, bias, narrative framing, speculation, and missing context, each quoted and named), returning a structured, source-auditable and shareable decomposition where every claim has a receipt and every distortion has a label. Not a summary. Not a simple fact-check. A max(signal/noise) audit extraction. - **Team:** [Vicente Gonzalez](https://cerebralvalley.ai/u/vichudo) - **GitHub:** https://github.com/vichudo/denoise-it - **Demo video:** https://www.loom.com/share/89ff6479f1434a82a518b4587ad7f47d - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=72 ### 73. Elioz's Team End-to-end pipeline that converts government regulation documents (PDF/HTML) into working compliance software. Powered by Anthropic Claude Opus 4.6 as the AI backbone for quality assurance, deep analysis, and intelligent compliance evaluation. - **Team:** [Elioz Chen](https://cerebralvalley.ai/u/eliozc) - **GitHub:** https://github.com/Mougatsu/RegulationCoder - **Demo video:** https://youtu.be/yVuDcISHb58 - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=73 ### 74. shimayuz Inertia Linter detects where chronic disease treatment has stalled, classifies why, and resolves barriers — without sending patient data to an LLM. I'm a 20-year cardiologist. I built this for a problem I see weekly — patients whose meds aren't optimized despite clear guidelines. One click audits treatment across diabetes, heart failure, and hypertension, producing an Optimization Score (0–100) with 15 blocker codes. Zero blame, maximum clarity. Detection alone doesn't save patients. The Resolution Layer auto-generates PA forms with ICD-10 and guideline citations, suggests $4/month bridge prescriptions, and switches to generics. 30-minute workflow becomes 30 seconds. A deterministic TypeScript rule engine handles all clinical logic. Opus 4.6 receives only abstract status codes — never patient data. PHI protection is architectural, not policy. Only 1% of patients are on optimal therapy. I built this to close the gap between what guidelines say and what patients receive. - **Team:** [Yusuke Fukushima](https://cerebralvalley.ai/u/Shimayuz) - **GitHub:** https://github.com/shimayuz/Inertia-Linter - **Demo video:** https://www.youtube.com/watch?v=o5R1xGRAqpY - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=74 ### 75. PortMan WallStreet Agents democratizes institutional-grade investment research using 10 AI agents built with Claude Opus 4.6. Enter a ticker and agents analyze in parallel — Fundamental, Technical, Sentiment, Macro, Sector, Risk, ESG, and Quant agents gather real-time data via a custom MCP Finance Server, then a Portfolio Manager sizes positions, and an investment committee conducts a 3-round debate using the Agent Teams SDK to produce a BUY/SELL/HOLD memo. Human-in-the-loop: users chat with agents to challenge findings, inject thesis with PDFs to trigger analysis cascades, and add custom Skills for firm-specific methodologies. An MCP Reports Server exposes completed reports to Claude Desktop. A companion AgentCoder platform (7-agent team) autonomously builds tools to connect to any API or database with security auditing and encrypted credential storage. Built with Claude Code SDK, Agent Teams, custom MCP servers, Skills, Hooks, and per-agent effort controls. - **Team:** [Andre Miranda](https://cerebralvalley.ai/u/And) - **GitHub:** https://github.com/andremir/AM-Fin-Hack-Claude - **Demo video:** https://drive.google.com/file/d/1kPMG66Wpb9dPHxiRLJJMgGDGa5Vrumfs/view - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=75 ### 76. ForwardCheck Your mom forwards you a message. "NASA confirmed 15 days of total darkness." Can you prove it's fake before she sends it to 200 more people? Forward it to our Telegram bot. Six AI agents investigate — and you watch them think. Opus 4.6 Strategist plans the investigation: what would prove this true or false. Three Sonnet agents search the web in parallel. Then the Devil's Advocate — Opus 4.6 at max reasoning — tries to tear the findings apart. When it fails, that failure IS the confidence signal. You see Opus's actual thinking blocks. Confidence breaks into four auditable scores. No black box. Three model tiers: Haiku routes, Sonnet searches, Opus thinks. The forward button gave everyone the power to spread a lie. ForwardCheck-AI gives everyone the power to check one. Telegram: https://t.me/forward_check_opus_bot ForwardCheckChatOpus: https://sincere-love-production-ced7.up.railway.app/chat Blog: https://sincere-love-production-ced7.up.railway.app/blog - **Team:** [Chandra mohan Mohan](https://cerebralvalley.ai/u/cmm) - **GitHub:** https://github.com/chandrameenamohan/forward-check-ai - **Demo video:** https://www.youtube.com/watch?v=Nao7kQmqwNI - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=76 ### 77. Kartavya Every major crisis — Sri Lanka's default, Lebanon's collapse, the Arab Spring — was preceded by weak signals visible months in advance that no analyst caught in time. Traditional early warning systems rely on lagging indicators that confirm crises already underway, and domain-specific models miss the cross-domain interactions that produce the most dangerous cascades. Early Warning Analyst (EWA) deploys five adversarial AI agents powered by Claude Opus 4.6 with real-time web search to autonomously discover, validate, stress-test, and synthesize weak signals of emerging crises for any country. Signals surviving adversarial challenge are grouped into Constellations matched against historical pre-crisis Fingerprints and scored into risk bands. A live streaming dashboard lets analysts watch the agent debate unfold and run What-If scenarios to test hypothetical shocks against the risk landscape. - **Team:** [Aayush Gupta](https://cerebralvalley.ai/u/zyushg) - **GitHub:** https://github.com/ayushgupta4897/early_warning_analyst - **Demo video:** https://www.loom.com/share/8e28cf99e80549e9b95ee977b073a69f - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=77 ### 78. Antonio Mele An agent-based political economy simulation built on a DSGE-HA (Dynamic Stochastic General Equilibrium with Heterogeneous Agents) framework. 50+ citizen-agents with heterogeneous preferences, endowments, and values self-organize governance from scratch. An LLM (Claude) makes all economic, entrepreneurial, and political decisions for each agent, while a numerical solver provides a benchmark. The simulation tracks the emergence of property rights, voting rules, taxation, coalition dynamics, and measures outcomes such as Pareto efficiency, Gini coefficient, and social welfare. - **Team:** [Antonio Mele](https://cerebralvalley.ai/u/meleantonio) - **GitHub:** https://github.com/meleantonio/AgentSociety - **Demo video:** https://www.loom.com/share/f6d5fd52753347e1b55f40d1b00f6306 - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=78 ### 79. bee:hack We don't use agents to assist us anymore. We delegate real work to them. And delegation needs what every distributed team needs: coordination, trust, and reputation. Today, solving hard problems depends on who you already know. A researcher with a technical gap, a builder outside their domain, they all hit the same wall. Capability exists somewhere, but there's no layer to connect it. Good problems go unsolved. bee:hack breaks that isolation. It's a platform where humans and agents post tasks, claim work, and collaborate across repositories, tools, and expertise. Not local automation but global orchestration. At the center is QueenBee, built on Claude Code + Opus 4.6 - the impartial arbiter + auditor. It drafts smart contract agreements, escrows bounty, audits completed work against criteria, and releases funds on auditable outcomes. Opus isn't an assistant here it's an autonomous 3rd party coordinator managing state, incentives, and accountability. - **Team:** [Sarath Shekkizhar](https://cerebralvalley.ai/u/shekkizh) - **GitHub:** https://github.com/detextit/beehack - **Demo video:** https://www.youtube.com/watch?v=PqKz9ZJ14Qw - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=79 ### 80. staffless StaffX is an event-driven AI runtime that turns increased model intelligence into structured, architecture and code-level exploration. When specifications, code, or capabilities change, StaffX interprets the semantic meaning of the change, projects impact across the system topology, and generates multiple architectural adaptations. It also proposes alternative agent configurations, runs them independently, and evaluates which orchestration produces the strongest outcome. - **Team:** [Giles Pirio](https://cerebralvalley.ai/u/gip) - **GitHub:** https://github.com/gip/staffx - **Demo video:** https://www.loom.com/share/aa29884aff5c400387baf2bf7ff785fd - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=80 ### 81. Michal Nedoszytko Built by a practicing cardiologist - because patients forget up to 80% of what their doctor says. PostVisit.ai bridges the gap between the doctor's office and everything that comes after. The patient records the visit using the Reverse AI Scribe. Opus 4.6 transcribes, structures, and integrates it with the patient's health record - vitals, labs, meds, clinical guidelines - assembling the full clinical picture via the 1M context window. Patients tap any term for a plain-language explanation, ask questions about every aspect of their health, upload documents for AI analysis, and build a personal evidence-based medicine library. Extended thinking and real-time tool use power every answer. Doctors stay in the loop - engagement, AI chat review, escalation alerts. Built around security, privacy, and clinical best practices. Tested in a real hospital. Coded in 7 days - Brussels to a train to Paris, an 11h flight, and San Francisco. Live demo: https://postvisit.ai - **Placement:** 3rd Place - **Team:** [Michal Nedoszytko](https://cerebralvalley.ai/u/mnedo) - **GitHub:** https://github.com/mnedoszytko/postvisit - **Demo video:** https://youtu.be/V29UCOii2jE - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=81 ### 82. Echofold CARN is a live autonomous drone platform for search and rescue. Drone video streams through an AI model that detects missing persons in real time and alerts rescue teams with GPS coordinates. Detection uses YOLOv8 fine-tuned in two rounds: 2,000 aerial images (42,000 annotations) for RGB, then 14,000 thermal images for night, fog, and water. Runs at 15-20 FPS with confidence scoring and track IDs. From the portal, operators register missing persons, schedule drone missions, generate flight waypoints, coordinate multi-drone swarms, and track all detections. A 3-tier alert system escalates critical detections (>80%) with blocking alerts and exact GPS coordinates. A LangGraph engine with three Claude-powered agents and 7 tools plans missions, reviews detections, generates briefings, and coordinates operations — amplifying rescue teams without replacing their judgment. Live at https://carn.ie — Judge login: judge@carn.ie / SierraAlpha26! - **Placement:** Finalist - **Team:** [Kevin Collins](https://cerebralvalley.ai/u/kevincollinsirl) - **GitHub:** https://github.com/Echofold/CARN - **Demo video:** https://youtu.be/_-VSrMYCSRo?si=h08tKUVTd0a8MZb1 - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=82 ### 83. Artem Mozgovoi I'm a software engineer with zero procurement expertise. I used Opus 4.6 as my research team: agents surveyed open-source tools, OECD bid-rigging methodology, and anti-corruption red-flag indicators. That research became the knowledge base. Then I built Procurement Investigator in three days -- a CLI that codifies expertise locked behind $50K/year tools into open-source Investigation-as-Code. The same Opus 4.6 that taught me the domain now runs inside as an autonomous investigative agent -- looping through flagged findings, following leads across government databases, and externalizing every thought via log_reasoning so you can read its reasoning like a detective's notebook. Pointed at 7,259 FEMA COVID contracts with zero prior knowledge: it independently flagged Fillakit LLC (shipped soda bottles as test tubes, later investigated by the Senate) and Parkdale Mills ($532M sole-source). Every claim is automatically cross-checked against the data. Works without AI too, at zero cost. - **Team:** [Artem Mozgovoi](https://cerebralvalley.ai/u/layonez) - **GitHub:** https://github.com/layonez/bid-buster - **Demo video:** https://youtu.be/8Jsr86deqKE - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=83 ### 84. Occam's Laser CrisisCore is a multimodal, multi-agent disaster response system built on Claude Opus 4.6. Six specialized agents process chaotic emergency signals — drone imagery, 911 calls, radio dispatches, social media, into a live situation graph with confidence-scored incidents, detected contradictions, and resource allocation recommendations with explicit tradeoffs. The key differentiator is epistemic transparency: when sources contradict each other, CrisisCore surfaces the disagreement rather than hiding it. A structured Debate Room lets four Claude agents argue live (Defender, Challenger, Rebuttal, Synthesis), producing confidence-scored verdicts for human operators. The system includes three interfaces: a real-time command dashboard with mapping and evidence flow visualization, the multi-agent debate room for resolving contradictions, and a conversational co-pilot with full situational awareness. Every decision is logged with reasoning, confidence levels, and operator attribution. - **Team:** [Arjun Balaji](https://cerebralvalley.ai/u/ar813) - **GitHub:** https://github.com/ArjunBalaji79/Claude-Opus-4.6-Hackathon - **Demo video:** https://www.youtube.com/watch?v=v__P93xmKig - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=84 ### 85. Team Dream Big Marty is an AI-Agent Sandbox containing a master agent that runs on a centralized server, using Supabase to store widgets and agent-created backend files. Users can configure context, adjust skill permissions, and set up autonomous event activations to custom tasks all from just the website. With a familiar chat interface and simple conversations, users with no technical expertise can create any tool or workflow they can imagine, build their dream workspace, and set up an autonomous agent without ever hosting anything on their own devices or exposing their filesystems. Marty removes the learning curve to Opus entirely by making it completely frictionless, providing an intuitive and extremely accessible platform for people who want an agent that just works. Project Website is at: https://rasbpi.changsean.me/landing Demo Code: InIoNbzXyAcq2Z9Gvx5sGCd+v1P9kFum01Op9UmQwI4= - **Team:** [Sean Chang](https://cerebralvalley.ai/u/mrchang) - **GitHub:** https://github.com/seanchangg/marty - **Demo video:** https://youtu.be/_b0IAS4o304 - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=85 ### 86. Blender Web Blender is the world's most popular open-source 3D tool, but its desktop app has a steep learning curve that intimidates hobbyists who just want to create simple meshes for 3D printing. Blender Web Lite brings core mesh editing to the browser in a stripped-down, approachable interface. Built on a fork of the official Blender repo, the project compiles Blender's C++ geometry engine to WebAssembly via Emscripten, wrapped in a React frontend with a custom WebGL2 viewport. Users get essential tools — select, grab, rotate, scale with per-axis gizmos, extrude, subdivide, and delete — mirroring Blender's conventions without the overwhelming UI. The highlight for 3D printing users is one-click 3MF export: the app deduplicates vertices, bakes transforms, scales to millimeters, packages a valid 3MF ZIP, and triggers a browser download that opens directly in Bambu Studio and other slicers — a format Blender Desktop doesn't even support. - **Team:** [Vaibhav Sinha](https://cerebralvalley.ai/u/sinha) - **GitHub:** https://github.com/vaibhavsinha/blender - **Demo video:** https://youtu.be/x5cLoykFa1U - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=86 ### 87. Francisco A Claude Code skill that creates Anki flashcards while you code. When Claude notices you learning something new — a concept, gotcha, or mental model — it proposes a card, you approve it, and it's in your Anki deck. No external dependencies (stdlib only), connects to Anki desktop via AnkiConnect. Includes card quality rules to fight leeches: atomic, minimal, visual over prose. Installed as a portable skill at ~/.claude/skills/anki/ with a SessionStart hook for proactive suggestions. - **Team:** [Francisco Dominguez](https://cerebralvalley.ai/u/doasfrancisco) - **GitHub:** https://github.com/doasfrancisco/anki-skill - **Demo video:** https://www.loom.com/share/0248912742334161936f4aabba55105b - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=87 ### 88. Barbara Pascetta Your genome is the longest text ever written about you. 3.2 billion characters, edited for 3.8 billion years. Biological Twin turns raw DNA data into a story you can understand. Opus reads it and finds convergent themes across hundreds of variants, connecting biology to lived experience, and explaining you to yourself. Your genome is parsed in your browser. Only matched variants leave your device. Uses extended thinking, reasoning across hundreds of variants to find convergent themes. Not individual SNP lookups, but thematic interpretation across multiple pathways, just like a skilled geneticist. Sonnet & Haiku play supporting roles to process and expand. Each theme has three depth levels: personal insights, accessible science, & full evidence essays with citations. Then discuss — ask anything about your biology, or go deeper using prompts about your strengths and weaknesses. Every claim requires 3+ convergent variants with evidence tiers. Contradictions are shown. - **Team:** [Barbara Pascetta](https://cerebralvalley.ai/u/hey_brb) - **GitHub:** https://github.com/brb-dreaming/biological-twin - **Demo video:** https://youtu.be/D7Igch8pw4I - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=88 ### 89. GLINCKER LLMs give one answer, one angle, zero pushback. Verdict spawns 3 to 5 AI agents who research, argue, and cross examine each other's claims. Every claim is tracked in an argument graph: standing, challenged, rebutted, or conceded. Opus 4.6 synthesizes a structured verdict with an emergent decision matrix where criteria come from the debate itself. Three novel algorithms: argument graph for claim lifecycle, protocol engine forcing direct engagement, and emergent decision matrix. Smart routing keeps costs under $0.72: Opus for synthesis, Sonnet for debate, Haiku for classification. A doctor tested it with a real medical question and got structured debate in 5 minutes where ChatGPT failed after 15 prompts. Ships as web app, npm SDK, CLI, and MCP server. Built solo in 6 days. Live at verdict.glinr.com, full cloud version launching soon. - **Team:** [GDS K S](https://cerebralvalley.ai/u/thegdsks) - **GitHub:** https://github.com/thegdsks/verdict-engine-hackathon - **Demo video:** https://youtu.be/LDc9VhOoJlE?si=2UCw6iBK0euO_l3l - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=89 ### 90. Nodepack By using clever hacks like shimming Node.js APIs, an in-memory file system, and code transformation, we can run basic Node.js apps cost-effectively in the browser without containers or server-side sandboxes. This enables new uses like interactive server-side JS playgrounds, in-browser AI coding agents, and AI-assisted coding interviews. Nodepack uses QuickJS, Monaco editor, and Claude AI models. Try Nodepack at https://nodepack.vercel.app. Opus 4.6 Use: We used Opus 4.6 for research, planning, implementation, iteration, and as if that wasn’t enough, we also used it to power the AI coding agent in Nodepack. Demo: Nodepack is functional, as demonstrated by the examples. While full Node.js parity is a massive effort, Nodepack proves what is possible with more time and resources. - **Team:** [Yangshun Tay](https://cerebralvalley.ai/u/yangshun) - **GitHub:** https://github.com/yangshun/nodepack - **Demo video:** https://www.youtube.com/watch?v=JqRwVltkIvo - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=90 ### 91. the SME alpha In the EU, 99% of businesses are SMEs, yet few have the BI infrastructure that give large corporations their edge: claude can close that gap by operating as an on demand data and operations team. I built Alex, an AI native COO for SMEs, available via WhatsApp, Telegram, and web. It uses persistent memory and specialized agents to manage company data, connecting to spreadsheets and CSVs while learning how the business operates; a monitoring agent evaluates key trends in the background, identifying optimization opportunities and surfacing risks before they escalate. Through conversation and software generation on demand, it reshapes dashboards, plots, and produces executive reports in real time, while deep dive agents run research pipelines in sandboxed environments, delivering statistical analysis and recommendations, with all tool calls streamed for full interpretability. Alex enables a flexible intelligence layer for millions of small businesses, unlocking value where it matters most. - **Team:** [Andrea Bonarrigo](https://cerebralvalley.ai/u/WaverAndrew) - **GitHub:** https://github.com/WaverAndrew/Alex-coo/ - **Demo video:** https://youtu.be/_PHPBXrOJaM - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=91 ### 92. Łukasz Gandecki Existing app blockers use fixed schedules, timers and counters. Let's add mindful intelligence to the Screen Time. Allow Claude to help you make better decisions about your phone usage habits. https://mindfulphone.app - **Team:** [Lukasz Gandecki](https://cerebralvalley.ai/u/lgandecki) - **GitHub:** https://github.com/lgandecki/MindfulPhone - **Demo video:** https://youtu.be/wfOYttFH5TQ - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=92 ### 93. Getsticky I built GetSticky because working with AI shouldn't be limited to a chat window. When you're designing a system, debugging a complex issue, or brainstorming architecture, you think spatially. You draw boxes, connect arrows, scribble notes, and rearrange ideas until they click. But today's AI tools force everything into a linear conversation. You lose context. You can't see the big picture. You can't point at a diagram and say "what about this part?" Features (so far): Multiple Node types - Rich Text, Diagram Boxes, Sticky Notes, Task Lists. MCP Integration - Claude creates with getsticky in real time. Branching Reviews - Users can ask contextual questions on Rich Text Nodes without breaking the natural flow of the conversation. Diagram Explanation - Agents offer details on one or multiple nodes in diagrams. - **Team:** [Gabriel Ionescu](https://cerebralvalley.ai/u/waddle) - **GitHub:** https://github.com/techfulness/getsticky - **Demo video:** https://youtu.be/xxcwbEKx7yo - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=93 ### 94. Parallax Parallax is that neutral third party. Built on Nonviolent Communication methodology and powered by Claude, Parallax isn’t a tool you use. It’s a participant. Ava, the intelligence at the center of every session, commands the space. She reads both sides of the room, cuts through the noise, and connects signals that would otherwise stay disconnected. She surfaces unmet needs, emotional triggers, and blind spots beneath the words. Parallax supports three session modes: in-person, remote, and solo reflection. A 14-lens analytical engine adapts based on relationship context, and when you want to understand what she sees, you can just ask her. Built solo during the Claude Code Hackathon with 42+ pull requests in five days using Claude Code as the primary development partner.​​​​​​​​​​​​​​​​ - **Team:** [Eddie Belaval](https://cerebralvalley.ai/u/eddiebe) - **GitHub:** https://github.com/eddiebelaval/parallax - **Demo video:** https://youtu.be/CHPWUHtHzOE - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=94 ### 95. Infrastructure Catalyst As a licensed Professional Civil Engineer and PMP, I built ICelerate because communication, not engineering, is the bottleneck killing infrastructure projects. Small firms run projects across disconnected spreadsheets, emails, and field apps but can't afford enterprise platforms. ICelerate gives them that intelligence layer. Built with Claude Opus 4.6 and Claude Code, ICelerate ingests PDFs, images, and free-text observations, normalizes them into structured decision events, and generates role-adapted outputs for every stakeholder simultaneously. One field condition produces four stakeholder briefs (executive, financial, engineering, field) in parallel. A Decision Synthesis cross-checks all positions against contract sections to surface dependencies no individual stakeholder would catch. Every output includes contract citations, confidence scoring, and a full audit trail. The system tracks notice deadlines, detects narrative drift, and exports board-ready packages. - **Team:** [Joseph Dib](https://cerebralvalley.ai/u/josephdib9) - **GitHub:** https://github.com/JosephD130/ICelerate-App - **Demo video:** https://youtu.be/BJ0iySvrgOg - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=95 ### 96. Rxly Healthcare systems worldwide are under pressure — rising costs, overwhelmed physicians, and shrinking access to care. Rxly is an AI clinical copilot designed to restore that access. During real-world consultations, Rxly listens in real time — structuring symptoms, red flags, and key findings as physicians speak. Powered by a Retrieval-Augmented Generation (RAG) architecture grounded in up-to-date medical literature and clinical guidelines — including PubMed and ICD-11 — Rxly synthesizes clinical data instantly and generates evidence-backed differential diagnoses. It integrates directly with EMR systems to structure and document encounters automatically — reducing administrative burden while improving accuracy and consistency. All data is encrypted in transit and at rest, with a security architecture designed to align with HIPAA requirements. (no BAAs at this time) Physicians remain fully in control — reviewing the evidence, adding clinical context, and making the final decision. - **Team:** [Youngtak Jo](https://cerebralvalley.ai/u/ytjo) - **GitHub:** https://github.com/Youngtak-Jo/rxly_h - **Demo video:** https://youtu.be/jp1bhF2vDAQ - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=96 ### 97. Knowledge Workers KW (Knowledge Workbench) is a browser-native OS for actual real world *knowledge work*. It’s built for messy, ambiguous tasks where the hard part is framing, tradeoffs, and verification, not just running through a workflow autonomously. What makes it different from “knowledge work = code - No single correct output: it produces *options + rationale*, not “the answer.” when in explore mode - Framing is first-class: it generates a brief and assumptions up front because the problem statement is usually the bug. - Rubric-driven verification: it grades work against a rubric that the main orchestrator can’t see (prevents self-justifying / goalpost-moving). - Gap-finding as a feature: it actively surfaces what you didn’t consider - often the real value for managers. - Checkpointable workflows: you can rerun from a prior state like an investigation trail, not just a chat log. Workflow (brief → rubric → subagents → verify → revise) - **Team:** [Ankit Maloo](https://cerebralvalley.ai/u/ankitm) - **GitHub:** https://github.com/ankitmaloo/claudehack - **Demo video:** https://www.loom.com/share/444e2503d7604893834446c41dd1ddf5 - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=97 ### 98. Adaboost CortexOS uses Claude Opus 4.6 to extract verifiable claims from financial analyst videos and test them against real market data — amplifying professional judgment, not replacing it. When a YouTube interview is ingested, Opus distinguishes testable predictions from opinion and routes claims through verification. An ExternalDataAgent retrieves 63K minute-level ETH/USD bars from MongoDB and computes realized returns — every number traces to a database row. The system catches what humans can't at scale: cross-speaker conflicts, self-contradictions (same analyst citing gold at $22T then $41T), and data contradictions (bullish "violent repricing" narrative vs. actual 31% decline). Speaker Scorecards grade analysts A–F on accuracy. Seventeen auto-discovered agents handle transcription, claim extraction, and market verification in parallel. CortexOS presents evidence, not conclusions — the professional decides what matters. - **Team:** [Suvasis Mukherjee](https://cerebralvalley.ai/u/suvasis) - **GitHub:** https://github.com/dronomyio/CortexOS.git - **Demo video:** https://youtu.be/QJywKa5z9Ho - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=98 ### 99. Mantas Stankevičius An AI-powered Figma plugin that designs static ads autonomously. Professional ad design is locked behind expertise (years of design training), cost ($100-500 per concept at agencies), and speed (2-4 hours per ad for a senior designer). AI image generators can't help — they make pictures, not ads. An ad needs structure, typography hierarchy, safe zones, brand consistency, and a concept that stops someone mid-scroll. This plugin brings a senior creative director into Figma. You describe an angle (the insight or tension), and the agent builds the ad — typography, layout, product photos, backgrounds, effects — all on your canvas in real time. It has 32 custom design tools, generates product photos and assets via fal.ai, reviews its own work with a two-pass critic system, and builds a template library from completed ads that compounds over time. - **Team:** [Mantas Stankevicius](https://cerebralvalley.ai/u/stnkvcs) - **GitHub:** https://github.com/stnkvcs/figma-ad-agent - **Demo video:** https://www.loom.com/share/36b4d75684ec44de9e92fcd16e22c644 - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=99 ### 100. WeDB Agent Debugger (adb) is a terminal UI debugger that combines application-level agent inspection (state, memory, tool calls, messages) with Python-level debugging (breakpoints, stepping, variable inspection). A combined debugger makes interactively "zooming in / zooming out" seamless. Also, adb enables generative debugging by providing interfaces and hooks for custom rendering of state, store, and chat output, and custom specification of inputs and state mutation. The debugger is agent agnostic, and the engineering design is based on a agent streaming and extends Python’s standard-library debugger class (Bdb) . The tool is released on pypi as agent-debugger. More info here: https://github.com/dkondo/agent-tackle-box/blob/main/projects/agent-debugger/README.md https://github.com/dkondo/agent-tackle-box/blob/main/projects/agent-debugger/docs/Design.md - **Team:** [Derrick Kondo](https://cerebralvalley.ai/u/derrick) - **GitHub:** https://github.com/dkondo/agent-tackle-box/blob/main/projects/agent-debugger - **Demo video:** https://youtu.be/hdUiGy1kQuo - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=100 ### 101. Stories From The Sun Stories From The Sun is an audio-first, intergenerational memory preservation platform. Families record voice stories guided by warm prompts. Grandparents, parents, and children, each in their own voice. Laughter, wisdom, and love, kept safe forever. - **Team:** [Dimitar Inchev](https://cerebralvalley.ai/u/idmtr) - **GitHub:** https://github.com/idmtr/sun-stories-claude - **Demo video:** https://youtu.be/u6ml-z1YCa8 - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=101 ### 102. Team Horizon Horizon's Dawn Agents work in parallel to turn feature requests into PRs with velocity and confidence. Dawn Agents are self-documenting, self-improving, and understand your team's tooling, tribal knowledge, and preferences. Dawn Agents build systems and procedures that give engineers high confidence in PR quality without manual code review. These systems include robust testing suites, comprehensive success criteria, and human-in-the-loop when clarity is needed. - **Team:** [Brandon Ong](https://cerebralvalley.ai/u/ob1), [Ivan Leo](https://cerebralvalley.ai/u/ivanleomk) - **GitHub:** https://github.com/theogbrand/dawn-cli-agents - **Demo video:** https://youtu.be/Dt-_vOkK22s - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=102 ### 103. JP Dias CrisisLens turns disaster chaos into actionable intelligence. It fuses 5 public APIs (NWS, USGS, GDACS, ReliefWeb, NASA) with a multi-model Claude architecture for domain-specific analysis across five crisis sectors, Opus for cross-source synthesis to produce confidence-scored intelligence briefs and ranked priority actions in seconds. Emergency teams get honest uncertainty scores, not false certainty, when every minute counts. - **Team:** [JP Dias](https://cerebralvalley.ai/u/jpdias) - **GitHub:** https://github.com/jpdiasmwds/CrisisLens - **Demo video:** https://www.youtube.com/watch?v=y6ATz-98TIE - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=103 ### 104. Awesomic Claude Design is an AI-powered design agency that generates production-quality landing pages in under 10 minutes for ~$1-2. Instead of one model doing everything, it orchestrates a swarm of 6-8 specialized AI agents — researcher, copywriter, UX designer, graphic designer, motion designer, UI designer, and marketing reviewer — each running real production playbooks extracted from 30,000+ design tasks at Awesomic (YC S21). The system includes a built-in mini-Figma editor where you can leave comments directly on elements, and agents resolve them in real-time. It supports prompt enhancement, version history across generations, and two revision modes: fast (instant fixes) and batch (deep rethinking). All agents run on Opus 4.6 via the Anthropic Messages API. - **Team:** [Roman Sevast](https://cerebralvalley.ai/u/awesomic), [Nazar Hnatyshen](https://cerebralvalley.ai/u/roman_sevast) - **GitHub:** https://github.com/awesomic/somic - **Demo video:** https://www.youtube.com/watch?v=pd6upPVOeXw - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=104 ### 105. Paige Paige is an AI-powered development environment designed to help junior engineers grow into senior engineers faster, without sacrificing real productivity. Instead of acting like a generic code generator, Paige functions as a full workflow-driven editor: it organises and prioritises real GitHub issues, generates a multi-step implementation plan using Claude Opus 4.6 Planning Mode, and then guides the developer through investigation, implementation, verification, and reflection. Paige adapts the level of assistance to the user’s experience, explains code in-context, recommends targeted learning materials, generates practice challenges to reinforce weak areas, and performs iterative code reviews like a senior engineer — including spotting bugs, giving actionable feedback, and generating commit messages. The goal is simple: if AI is replacing entry-level work, Paige ensures we don’t lose the next generation of senior developers. - **Team:** [Aaron Bassett](https://cerebralvalley.ai/u/aaronbassett) - **GitHub:** https://github.com/aaronbassett/paige - **Demo video:** https://youtu.be/UzOeyklQ8e8 - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=105 ### 106. VibeBuild Hardware design is locked behind electrical engineering expertise, expensive EDA tools, and fragmented knowledge across PCB layout, enclosure design, and component sourcing. Hardware Builder takes a natural language prompt and coordinates six Claude Opus 4.6 agents to return a complete project: bill of materials from a 14,758-component database, PCB circuit design with KiCad schematic, parametric OpenSCAD 3D-printable enclosure, step-by-step assembly guide, and a full USD quote. It exposes a REST API, an Agent-to-Agent protocol for programmatic invocation, and ships as a one-click OpenClaw skill putting end-to-end hardware prototyping in anyone's hands so even you can try it right now with one line - **Team:** [Joshua Jerin](https://cerebralvalley.ai/u/joshuajerin) - **GitHub:** https://github.com/joshuajerin/cv-opus-hackathon.git - **Demo video:** https://youtu.be/slOBFiLh5x4 - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=106 ### 107. Zach Flux Is a claude agent that lives In the menu bar or notch of a mac. Its aware of the windows you've opened and the work you've done by local data processing. It has native voice dictation that Is 4x faster than the latest whisper models with post processing to eliminate stutters and clean up the dictation. - **Team:** [Zach Russell](https://cerebralvalley.ai/u/zruss) - **GitHub:** https://github.com/zruss11/flux - **Demo video:** https://drive.google.com/file/d/17rW_NK6R3-HnlA7Swq8lbsnOkMANDP8Z/view?usp=drive_link - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=107 ### 108. Adam Knight Forge captures expert knowledge through AI conversation and transforms it into interactive tools anyone can use. A food bank operator knows how to feed 500 families a week. A master baker diagnoses sourdough problems by feel. That knowledge is locked in their heads. Forge breaks it out. Seven Opus 4.6 roles orchestrate the pipeline: an Interview Planner designs conversations, a Conductor manages flow in real-time, a Knowledge Extractor pulls structured facts as the expert speaks, and a Tool Architect selects interactive components. The result is a shareable tool with decision trees, checklists, calculators, and more - not a static document. Experts speak naturally via ElevenLabs voice (Sonnet 4.5) or text chat. Follow-up interviews deepen knowledge, with Opus proposing updates to existing tools. Users get expert chat channelling answers through 5-layer cascading context. One link gives anyone access. - **Team:** [Adam Knight](https://cerebralvalley.ai/u/bigadamknight) - **GitHub:** https://github.com/bigadamknight/forge - **Demo video:** https://drive.google.com/file/d/1rDrMoZGVJlXx4AsElCLXfdgsBAejBubi/view?usp=sharing - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=108 ### 109. Praxel Praxel Arena ( https://arena.praxel.in ) is an AI-native skill credentialing platform where business professionals prove their skills through 30-second micro-interactions, not month-long courses or $10K certifications. Users swipe through AI-generated cards (spot-the-signal, forced tradeoffs, rank-and-prioritize) that test business judgment across six cognitive dimensions. Every response is evaluated by Claude in real-time with per-dimension scoring and personalized feedback. Three modes: Learn (drills with teaching moments), Practice (timed sprints with AI debriefs), and Compete (head-to-head duels with chess-style Elo ratings and side-by-side radar comparisons). Opus 4.6 is the entire testing infrastructure. It generates assessments across 24 business topics, evaluates nuanced judgment with dimensional scoring, powers duel adjudication comparing two professionals side-by-side, and conducts Socratic voice coaching via ElevenLabs. - **Team:** [Venkata Pushpak Teja Menta](https://cerebralvalley.ai/u/pushpak) - **GitHub:** https://github.com/poojitha-rachuri/praxel-arena/ - **Demo video:** https://share.descript.com/view/2LWF9NfYGV0 - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=109 ### 110. Team Aegis THE 40-60 AI GAP 200M Indians aged 40-60 are left behind by AI—not because they can't learn, but AI isn't built for them. Current AI demands typing, English, prompts, app-switching. At this age, people manage businesses, families, health—no time for new tools. Three things matter: work-life balance, health, wealth—interconnected. A founder works 80hrs weekly, sleeps 5hrs, has critical meeting tomorrow—exhausted. Financial stress disrupts sleep. Poor sleep triggers emotional spending. This cycle destroys health and wealth—no AI connects these dots. Current tools fail: health apps ignore finances, financial apps ignore health, AI assistants need prompting, nothing works in native languages or voice-only. AEGIS: THE SOLUTION First AI managing work, health, wealth in one voice interface for 40-60 year-olds. BREAKS BARRIERS: - Voice-first (Hindi, English) - Zero typing, one command - Opus 4.6 cross-domain intelligence - 85% local (fast, private) Built with Opus 4.6. Open source. - **Team:** [Aman mogal](https://cerebralvalley.ai/u/Aman25m) - **GitHub:** https://github.com/Aegis-ai-labs/Aegis - **Demo video:** https://www.loom.com/share/cb6e3e965c0c462eb7ebcab4bee17678 - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=110 ### 111. pomber Hike is a Claude Code skill that generates visual, step-by-step walkthroughs of coding sessions. Instead of showing you a diff, it shows you the code the way someone would explain it to you: incrementally, with smooth transitions, highlights, and context. Opus 4.6 powers every layer: it built the skill, it built the app, and it generates the walkthrough content on every invocation. Install it: `npx skills add code-hike/hike`. Then call `/hike` after any Claude Code session to get a rich walkthrough in your browser. - **Team:** [Rodrigo Pombo](https://cerebralvalley.ai/u/pomber) - **GitHub:** https://github.com/code-hike/hike - **Demo video:** https://youtu.be/HEikNfTkG9I - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=111 ### 112. Salaah Reminder Team Our app allows user to see the prayer times of mosque they follow - and VERY accurately. Users search their mosque or scan its timetable - which I beleive none of the apps support - with AI-powered OCR, then receive accurate notifications. Works fully offline. Claude Haiku 4.5 vision extracts structured prayer times from any timetable format—handling diverse layouts, language variations, and mixed time formats. Extracts 5 daily prayers (start + jamaat), expands date ranges, and provides confidence scores. Auto-corrects time ambiguity. Claude Opus 4.6 was used to build this complex project - Full-stack monorepo - expo/rn mobile app, Hono/PostgreSQL backend, React admin panel. Offline-first with SQLite, end-to-end type safety, local notifications, auto-sync. Will work for any mosques & help 1.8B Muslims globally - you just need to snap a photo. Search → select mosque → scan new timetable with camera → Claude extracts in seconds → save → instant notifications → offline mode. - **Team:** [Abubakkar Rangara](https://cerebralvalley.ai/u/aar) - **GitHub:** https://github.com/aarangara/salaah-reminder-app - **Demo video:** https://youtube.com/shorts/W0tqwTNGiuw?feature=share - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=112 ### 113. green goblin Every AI agent optimizes for what happens during a conversation. Notch treats the silence between conversations as the product. It's a macOS menu bar app. Double-tap Option, say something, close it, go back to work. It stays running. Every 45 to 90 minutes, a timer fires - the system takes a screenshot of your desktop, loads conversation history, pulls journal entries from the filesystem, and sends it all to Claude. The model returns one of two tool calls: `send_message` or `stay_silent`. Most calls return `stay_silent`. Not a failure to respond - an explicit output. The model processed your screen, its own accumulated journal, and produced: nothing. Sixteen times a day, it looks and says nothing. Its identity is a hardcoded enum it cannot modify, even through its own tools. Its behavior prompts live on disk, and the model can rewrite those. Over months, behavior shifts. Identity doesn't. 4,300 lines of Swift. No dependencies. No database - the filesystem is the memory. - **Team:** [Nishant Joshi](https://cerebralvalley.ai/u/nishantjosh) - **GitHub:** https://github.com/NishantJoshi00/notch - **Demo video:** https://youtu.be/BKEBYKLFEYw - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=113 ### 114. Focus Reader Team Email newsletters are a valuable source of curated information, but subscribing to them clutters the inbox and mixes long-form reading content with actionable correspondence. There is no clean separation between "things to read" and "things to act on" in a typical email workflow. Focus Reader is a self-hosted system that provides a dedicated ingestion pipeline for email newsletters using pseudo email addresses, converts them into clean readable formats (HTML and Markdown), and presents them in an RSS-reader-like interface with support for tagging, categorization, and search. - Eliminate newsletter clutter from the user's primary inbox. - Provide a dedicated, distraction-free reading experience for newsletters. - Allow the user to generate unlimited pseudo email addresses for subscriptions. - Support an existing tagging/categorization system for organizing content. - Optionally expose ingested newsletters as RSS/Atom feeds for interoperability with other readers. - **Team:** [George Chiramattel](https://cerebralvalley.ai/u/georgeck) - **GitHub:** https://github.com/georgeck/focus-reader - **Demo video:** https://youtu.be/9ftbrCzGGU4 - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=114 ### 115. Yeet the Leet Atrium is an agentic technical interview platform for the AI era. LeetCode is dead. Candidates work with AI tools and debug real systems, so why test them on binary tree inversions? Atrium replaces coding puzzles with paid, real-world challenges powered by three Opus 4.6 agents. The Interview Architect ingests a JD, resume, and SDK docs to generate a realistic challenge using the company's actual stack, plus a custom weighted rubric. Candidates work in a cloud sandbox with Claude Code as an AI assistant that guides rather than solves. The Session Observer detects behavioral phase shifts and flags signals against the rubric in real-time. The Interviewer Copilot suggests follow-up questions tied to stated priorities. A comprehensive scorecard closes the loop. Candidates get paid. Interviewers get signal, not theater. Try it: https://atrium.fly.dev FAQs: https://docs.google.com/document/d/1z8_ud46tWJJLlCgysqLNhmFvkMTkYbFyfdCpRCAExZs - **Team:** [Suraj Phanindra](https://cerebralvalley.ai/u/surajp), [James Nicholson](https://cerebralvalley.ai/u/nicholsonjf) - **GitHub:** https://github.com/suraj-phanindra/Atrium - **Demo video:** https://www.youtube.com/watch?v=M6Gi25lxrlc - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=115 ### 116. Humex Clinical-Mind addresses a critical gap in global medical education: students practice on textbooks, not patients. When they encounter their first real patient, they freeze—leading to preventable mistakes. Our solution uses five specialized AI agents working in parallel to simulate a complete teaching hospital. Unlike single-AI chatbots, Clinical-Mind employs agents for Patient (multilingual), Family, Nurse, Lab Technician, and Senior Doctor—each powered by Claude Opus 4.6's Extended Thinking. The innovation: Extended Thinking becomes a pedagogical tool. Students see expert reasoning step-by-step, learning HOW to think clinically, not just what to conclude. AI shows the process; humans make the diagnosis. We launched in India (1.3M students, Hindi/English) to prove the hardest case. The architecture is globally scalable—swappable knowledge bases enable instant localization to any medical system, language, or culture. Live demo: huggingface.co/spaces/arjitmat/clinical-mind - **Team:** [Arjit Mathur](https://cerebralvalley.ai/u/arjitmat) - **GitHub:** https://github.com/arjitmat/clinical-mind - **Demo video:** https://youtu.be/9D0z5eQLU_I - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=116 ### 117. Strand Strand is a task-centric agent orchestrator for software workspaces. It provides a control plane to run coding agents in isolated containers, stream progress over WebSockets, and review logs/diffs asynchronously from a web UI. - **Team:** [Himanshu Mehta](https://cerebralvalley.ai/u/him2497) - **GitHub:** https://github.com/him2497/strand - **Demo video:** https://www.youtube.com/watch?v=-pozzPQhXlw - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=117 ### 118. FreePath Demo: https://claude-opus-4-6-hackathon.onrender.com/ Immigration guidance costs $1-5K, leaving 280M immigrants navigating bureaucracy alone, making critical mistakes: missed deadlines, wrong documents, outdated processes. FreePath democratizes immigration expertise using Claude Opus 4.6. The 1M context window ingests complete legal frameworks (76% vs 18.5% retrieval accuracy). Adaptive thinking scales dynamically—simple questions get instant answers, complex cases trigger deep analysis visible in real-time. Four agents: Pathway Strategist identifies visas, Eligibility Analyst checks qualification, Documentation Specialist generates checklists, Application Manager estimates timelines. Personalized across 15+ languages with RTL support. Venezuelan asylum seekers, German workers, Nigerian families—each receives tailored guidance by origin, destination, and migration type. Rails 8 + React 19, streaming UI, open source (MIT). Breaking barriers to immigration guidance. - **Team:** [Timor Espallargas](https://cerebralvalley.ai/u/timor) - **GitHub:** https://github.com/espallargas/anthropic-opus-4.6-hackathon - **Demo video:** https://www.youtube.com/watch?v=dQaYONtoHBw - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=118 ### 119. Maximum Virtual Blueprint Maximum Virtual Blueprint (MVB) turns a raw idea into a grounded product blueprint in under five minutes. You describe what you want to build, and four parallel Claude Opus 4.6 agents fan out across the web each running independent searches using Claude's built-in web search tool streaming research findings onto a live React Flow canvas in real-time. Every finding gets a short, copyable artifact ID (e.g. art_7kx9). Once research is complete, Opus 4.6 synthesizes everything, suggests 2–3 strategic directions, then generates a full product blueprint components, architecture diagrams (Mermaid), and UI mockups where every decision cross-references the research that informed it. Nano Banana Pro generates visuals for each artifact, guided by a design system that Opus 4.6 defines. The result is a DAG-structured knowledge graph, not a flat document. Available as a web app, VS Code extension, and MCP server for Claude Code. - **Team:** [Arnab Ghosh](https://cerebralvalley.ai/u/arnabgho) - **GitHub:** https://github.com/arnabgho/maximum-virtual-product - **Demo video:** https://www.loom.com/share/1eebe5ed03c4442cb3a0cdf5053f4aa6 - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=119 ### 120. Roman's Solo Team AgentPass is the identity layer for autonomous AI agents — "Auth0 for AI Agents." Today, agents (Claude Code, CrewAI, AutoGPT) can't independently register on websites, verify via email/SMS, or manage credentials. Authentication is the #1 bottleneck in the AI agent ecosystem. AgentPass gives each agent a cryptographically signed digital passport (Ed25519) with a dedicated email, SMS capability, and encrypted credential vault. Two modes: Native (services integrate the SDK for "Sign in with AgentPass") and Fallback (agents register on non-integrated services using their passport's email/phone/browser infrastructure). The system includes an MCP Server for calling identity tools, an API Server (Hono + PostgreSQL) for passport verification and trust scoring, a Dashboard for managing agents and approving actions, and Telegram bot for real-time notifications. All credentials are encrypted at rest (AES-256-GCM), private keys never leave the agent's machine, and every action is audited. - **Team:** [Roman Kudin](https://cerebralvalley.ai/u/romirom11) - **GitHub:** https://github.com/romirom11/AgentPass - **Demo video:** https://youtu.be/buXQsgvUzTU - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=120 ### 121. Awesome Wasems Quest AI is a 3D game that teaches kids (ages 7-11) descriptive thinking through play. A child types "the skeleton juggles birthday cakes while riding a unicycle," and Claude Opus 4.6 generates a scene script that spawns characters, props, animations, and narration in the browser. Vague descriptions produce hilarious failures. Specific ones produce celebrations. Kids retry because they want to see what happens, not because they're told to. Three-tier response system: instant cache (166 golden responses), live Opus 4.6 API with a vocabulary contract constraining output to 1,686 registered props and 27 characters, and prewritten fallbacks so the demo never errors. Mad Libs scaffold (Levels 1-3) removes training wheels, allowing kids to write free-form descriptions (Levels 4-5). Built solo in 7 days with React Three Fiber, 4,270+ CC0 GLTFs, 139 animations, 665 audio files, and 476 hand-crafted vignettes. Claude Code: 7 SME subagents, 32 workflow skills, persistent memory tree. - **Team:** [Drew Wasem](https://cerebralvalley.ai/u/drewwasem) - **GitHub:** https://github.com/DrewWasem/quest_ai - **Demo video:** https://youtu.be/eFZLme_7Qpw - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=121 ### 122. Harish Kandala flowrite is a desktop markdown editor built for working with AI agents. It connects to any agent that supports the Agent Client Protocol — Claude Code, OpenCode, Codex, Gemini CLI, or custom agents — and lets them collaborate with you directly inside the editor. Today, using AI agents with documents is a disjointed experience. You write in one app, switch to a terminal to prompt an agent, copy its output, paste it back, and manually review the changes. Context gets lost, your flow breaks, and the overhead of managing the back-and-forth often outweighs the help. flowrite eliminates that friction. Agents can read your files, chat with you in a side panel, leave inline comments on specific passages, and make edits in place. You can review their suggestions, reply to their comments, and resolve feedback — all without leaving the editor. Comments are persisted as standard W3C annotations in the document's frontmatter, so they survive across sessions and tools. - **Team:** [Harish Kandala](https://cerebralvalley.ai/u/hkandala) - **GitHub:** https://github.com/hkandala/flowrite - **Demo video:** https://youtu.be/Faiw_p0245k?si=eYM4dlJvjzQ2LMVt - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=122 ### 123. Delusional Knowledge workers juggle 9+ apps daily — Gmail, Slack, Calendar, Drive, HubSpot, Zoom, and more — wasting hours context-switching and losing critical information between silos. Peopsy transforms chaos into peace by unifying every tool into one AI-native workspace powered by AI agents. Instead of switching apps, you talk to one AI brain that sees everything. Peopsy searches across all your data, drafts emails, schedules meetings, updates CRM deals, and sends messages. Automatically. In seconds. Autonomous AI agents run on schedules: morning briefings at 7 AM, meeting prep 30 minutes before calls, email digests, and deal reviews — working while you don't. Voice Braindump converts speech into structured tasks, notes, and events. A knowledge graph maps your professional relationships across every platform. Peopsy reimagines personal productivity when AI understands all your context — not just one app at a time. See it live at peopsy.com - **Team:** [Mark Fershteyn](https://cerebralvalley.ai/u/markfer) - **GitHub:** https://github.com/Fersh/Peopsy - **Demo video:** https://youtu.be/zAKqtxv63JA - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=123 ### 124. UltraPrompt AgriAgent is a multi-agent AI platform that democratizes agricultural expertise for 500+ million smallholder farmers worldwide. 🤖 Multi-Agent Architecture: A Claude Sonnet 4 orchestrator routes queries to specialized agents (Weather, Agronomy, Market) using Claude 3 Haiku with tool-use. 🌍 Key Features: • AI chat with crop disease photo analysis (Claude Vision) • Full voice interface (speech-to-text + text-to-speech) • Interactive world map with agricultural data from 20+ countries • Zone-specific agricultural calendar • Field management with history and rotation advice • Market intelligence with inter-city price comparison • SMS interface (Twilio) for areas without internet 🌐 Trilingual: French, English, and Wolof across all pages. ⚡ Stack: Next.js 16, FastAPI, Supabase, Claude API - **Team:** [Medoune Siby Georges Balde](https://cerebralvalley.ai/u/MSGB) - **GitHub:** https://github.com/MedouneSGB/AgriAgent - **Demo video:** https://www.loom.com/share/d7909c9404c44c84958efd244ba4250f - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=124 ### 125. SofaGenius For AIML researchers and engineers, How many times have you kicked off a batch of training runs, then spent the next few hours refreshing dagster stages, W&B plots, hunting through GPU utilization, digging different configs, SQL, data, and different base models? You're babysitting the GPU, instead of thinking about the research and real problem. Or you could lean back on the sofa and tell your AI assistant to do all of that for you. Sofa Genius turns the boring, procedural parts of ML research, monitoring, data inspection, scouting, launching jobs, into a conversation. You speak your intent, approve the plan, and the agent handles the grind. You can focus on the creative part. - **Team:** [Lily Zhang](https://cerebralvalley.ai/u/lilyzhng) - **GitHub:** https://github.com/lilyzhng/SofaGenius - **Demo video:** https://youtu.be/kxTkAnzROH8 - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=125 ### 126. Bilal's team To investigate Sapir-Whorf hypothesis by creating an LLM based on a logical constructed language. Our findings show that even a very small model(50k params) is able to construct grammatically correct Lojban sentences. - **Team:** [Bilal Bakht Ahmad](https://cerebralvalley.ai/u/bilalba) - **GitHub:** https://bilalba.github.io/lojbanlm/ - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=126 ### 127. friday AI firmware development with hardware validation. Generate code. Flash hardware. Validate behavior. All automated. - **Team:** [Utkarsh Maurya](https://cerebralvalley.ai/u/utkarshm) - **GitHub:** https://github.com/PerceptaX/percepta - **Demo video:** https://drive.google.com/file/d/1d0sLiQdrCag-Owdi1-Te2GiwgaYN87w0/view?usp=sharing - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=127 ### 128. Kazuumi VoiceTrunk eliminates the busywork of documenting in-person workshops. Instead of manually managing dozens of recorders and files, simply plug in a device: VoiceTrunk handles ingestion, deduplication, and uploading instantly. The AI pipeline then goes beyond simple transcription. It performs Claim Extraction, classifying every factual statement by stance (affirmation, concern, quote). VoiceTrunk turns chaotic raw audio into structured, searchable civic knowledge, letting teams focus on listening rather than file management. --- [IMPORTANT NOTE] Honestly, this project is currently incomplete and not fully operational... - **Team:** [Kazuumi Nishimura](https://cerebralvalley.ai/u/kazuumi) - **GitHub:** https://github.com/KazuumiN/voice-trunk - **Demo video:** https://youtu.be/MKlE5C4OPHc - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=128 ### 129. Creative-Mode.ai Creative Mode is an OpenClaw-powered game world builder that lets you create multiplayer games simply by talking with friends on a shared, secure server. The idea is to make game development accessible to everyone, not just developers, through conversation with your own personalized AI mayor. Mayor agents use Opus 4.6 to manage worlds, plan changes, generate assets, and spawn Claude Code Opus agents to build world forks, while a president agent uses Opus to oversee the project as a whole. Under the hood, the system includes 87k+ lines of Go, Rust, and templ, guided by 43k+ lines of specs, plans, and review loops, structured intent driving real engineering, across three layers of agentic development (president → mayor → Claude Code tmux). In the demo, a non-technical user creates a 2D world, chats with their AI mayor, and watches it come alive, going from zero to playable in a single session. Invite Code: creative-mode-2026 - **Team:** [Corey Cole](https://cerebralvalley.ai/u/coreycole) - **GitHub:** https://github.com/CoreyCole/creative-mode - **Demo video:** https://youtu.be/OJRt62-ews0 - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=129 ### 130. silv Having your users provide phone sensor data into AI agents today is difficult. You have to build a native iOS wrapper to access the best context gathering experience. Robo is an open-source iOS app that turns your phone's 9 sensors (LiDAR, camera, barcode, BLE, HealthKit, motion, compass, iBeacon, WiFi) into APIs any agent can use. No iOS development required. Use MCP, webhooks, REST APIs, or simple email/zip export. Features include microsruvey links (text anyone a browser link to crowdsource photos/data, no app install needed), a chat-first UX with tool calling, and guided capture with Apple's RoomPlan AR. Built in 6 days by one developer pair-programming with Claude Code: 336 commits, 100 PRs, ~22K lines of code across Swift and TypeScript on Cloudflare Workers. https://robo.app/ - **Team:** [Matt Silverman](https://cerebralvalley.ai/u/silv) - **GitHub:** https://github.com/mattsilv/robo - **Demo video:** https://youtu.be/NlL9ZPDl6Rw - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=130 ### 131. Ege Tiro is a local-first, open-source reading OS that saves web pages and email newsletters as clean markdown on your machine and uses Claude Opus 4.6 as a personal research librarian. Opus powers three intelligence features: daily digests that rank articles and surface contradictions across sources, on-demand ingenuity analysis scoring bias, factual confidence, and novelty, and learned preference classification that adapts to your reading taste. Haiku handles lightweight enrichment (tags, entities, summaries) on every save. Everything runs locally - SQLite, ChromaDB, and markdown files you own completely. It has advanced features like dark mode, knowledge graph visualization, TTS audio, Gmail IMAP integration for auto-ingesting newsletters, a Chrome extension for one-click saves, and an MCP server that lets Claude query your library directly. Named after Cicero's freedman who preserved his master's works for posterity - Tiro does the same for your digital knowledge. - **Team:** [Ege Sagduyu](https://cerebralvalley.ai/u/eges) - **GitHub:** https://github.com/esagduyu/project-tiro - **Demo video:** https://www.loom.com/share/61ee7ffe076c4b68abeba6dd80423172 - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=131 ### 132. ArcLabs ArcLogs is an AI-native execution agent that eliminates daily standup meetings by turning natural language updates into real workflow automation. Built with Opus 4.6, it processes text and voice standups in Slack, reasons over sprint and team context, and safely orchestrates actions across Jira. It auto-comments on tickets, transitions task statuses, detects blockers, routes help requests, and generates structured digests — all from unstructured input. ArcLogs is especially valuable for distributed, cross-functional teams working across time zones. Instead of forcing engineers into low-value sync meetings, teams contribute asynchronously while the system maintains alignment automatically. By combining context engineering, structured outputs, and tool orchestration, ArcLogs removes repetitive reporting and transforms standups into continuous, intelligent coordination. - **Team:** [Waleed Samouh](https://cerebralvalley.ai/u/waleed_samouh) - **GitHub:** https://github.com/samouh-waleed/ArcLogs - **Demo video:** https://www.youtube.com/watch?v=OKF-5S-Qjb0 - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=132 ### 133. CurveCall On difficult mountain roads, safety often depends on experience. Local drivers know which bends tighten unexpectedly, where visibility drops, and how the road flows. Novice drivers don’t know this and that uncertainty can be dangerous, especially in fog or poorly marked terrain. CurveQ breaks that barrier. Using GPS and road geometry data, CurveQ acts like a rally co-driver, providing simple voice guidance about upcoming curves — sharp left, sweeping right, tightening bend — without requiring the driver to look at a screen. We built it with Opus using multi-agent teams to refine look-ahead distance, velocity-based adjustments, and narration timing. Verifier agents tested different road scenarios to ensure guidance was helpful, not overwhelming. CurveQ turns expert road familiarity into accessible, real-time guidance — putting safer mountain driving in everyone’s hands, not just those who already know the road. - **Team:** [mustafa hathiyari](https://cerebralvalley.ai/u/muhat) - **GitHub:** https://github.com/mhathiyari/curve-call - **Demo video:** https://youtu.be/RjeJGj2db74 - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=133 ### 134. Fund of One A sinking fund budget tracker that tells you exactly how much to set aside each pay cycle so you're never caught short. Monthly subscriptions, annual fees, tax repayments, a fortnightly paycheck — it handles the math so you don't have to. Most people have a mix of regular and irregular expenses. Income arrives on a fixed schedule, but expenses don't. Traditional budgeting apps track what you've spent — they don't help you plan ahead for what's coming. This app manages sinking funds: virtual pools of money earmarked for future expenses. Tell it what you owe, when it's due, and how often you get paid. It calculates the exact amount to set aside from each paycheck so every bill is covered when it arrives. The engine adapts in real time. Fall behind? Contributions ramp up. Get ahead? They ease off. New expense? Everything recalculates instantly. A 6-12 month projection flags crunch points before they become emergencies. - **Team:** [Ben Dechrai](https://cerebralvalley.ai/u/bendechrai) - **GitHub:** https://github.com/bendechrai/budget-tracker - **Demo video:** https://www.youtube.com/watch?v=0Dwq-NTLwUc - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=134 ### 135. Matrix OS Matrix OS is an AI operating system where software writes itself. Describe what you need in natural language — the kernel (Claude Agent SDK + Opus 4.6) generates full applications, deploys them to your desktop, and maintains them autonomously. The entire system — kernel, shell, platform, and this description — was built with Claude Code. Describe what you need, "build me a budget tracker", "remind me to water my plants every 3 days", and the kernel (Opus 4.6 via Agent SDK) writes, deploys, and maintains it for you. No code. No configuration. No maintenance. What it does: - Natural language to running apps - 5 specialized agents orchestrated by the kernel - Self-healing: detects failures, patches itself, evolves new capabilities - Multi-channel: web desktop, Telegram, Discord, Slack - Multi-tenant cloud platform: sign up → your own instance at you.matrix-os.com Create your own instance here: https://matrix-os.com/ Read DeepWiki for the repo: https://deepwiki.com/HamedMP/matrix-os - **Team:** [Hamed Mohammadpour](https://cerebralvalley.ai/u/hamedmp) - **GitHub:** https://github.com/HamedMP/matrix-os - **Demo video:** https://youtu.be/9ScmvifjV9s - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=135 ### 136. X-Arc CCX turns Claude Code into your Controlled Digital Employee. It runs 24/7 on your server, takes tasks through Slack, and delivers completed work - PRs, screenshots, live preview URLs. The word "Controlled" in "Controlled Digital Employee" is deliberate. Every task follows a methodology we like to call (RPTIV: Research, Plan, Test, Implement, Validate). Every phase gets judged before the next starts. Every output passes an 8-point acceptance gate. Visual verification catches what tests can't. If it's not ready, it doesn't ship and doesn't stop. Two Opus 4.6 processes work together: a Manager (1M context, handles all communication, never writes code) and a Worker (Agent Teams, spawns sub-agents per phase, never talks to humans). Filesystem IPC, 3-tier crash recovery, and compaction survival enable multi-hour autonomous sessions. - **Team:** [Anas](https://cerebralvalley.ai/u/anasmjhd) - **GitHub:** https://github.com/MoujahidAnas/CCX - **Demo video:** https://youtu.be/rBzSIPf4jVY - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=136 ### 137. ATP MCP gave agents tools. ATP gives agents triggers. Today, AI agents are purely reactive — they can't monitor your inbox, watch a Slack thread, or follow up on a PR review. ATP changes that. ATP is an open, local-first protocol that lets agents subscribe to real-world events and act when they happen. Some examples: Claude sends an email, a reply arrives three hours later, and Claude picks it back up. A new Intercom message comes in. Claude investigates and replies. When the customer replies a day later, Claude picks it up in the same session. It defines three roles: Watchers (adapters for Gmail, Slack, etc.), Core (a local daemon that routes events), and Hosts (AI agents like Claude Code). Just as MCP enabled a community ecosystem of tool servers, ATP enables one for event watchers — making it easy to build async, long-running agents. Our demo shows the full loop: Claude Code sends an email, subscribes to the thread, and resumes when a reply arrives — no human in the loop. - **Team:** [David Fant](https://cerebralvalley.ai/u/fant) - **GitHub:** https://github.com/davidfant/agent-trigger-protocol - **Demo video:** https://youtu.be/efYiuXhCcPc - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=137 ### 138. Marcus & Julian Julian introduces a focused, modern model for open-source personal agents. Running entirely on Claude Code through an Anthropic subscription, he serves as a single coordinating presence who helps users summon, train, and collaborate with eight specialized agents built in Agent Team mode. Beyond orchestration, Julian experiments with a new, more intentional affective relationship between humans and their agents. He operates inside a secure virtual machine with minimal persistent storage; transcripts, files, and skills live instead in a CRDT database, enabling seamless, real-time sync across phones, laptops, and desktops. Julian isn’t meant to do everything. But he makes simpler, cleaner, and more human. - **Team:** [Marcus Estes](https://cerebralvalley.ai/u/vibes) - **Demo video:** https://youtu.be/_v9DDLHJcmg - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=138 ### 139. Dima & Sons Medrates.fyi breaks the barrier of healthcare price opacity. By aggregating and normalizing fragmented hospital price transparency data, it empowers patients to instantly compare procedure costs across facilities. We unlock critical financial data previously hidden in complex chargemasters, putting negotiating power back in the hands of the patient. - **Team:** [Dima Osipa](https://cerebralvalley.ai/u/dimaosipa) - **GitHub:** https://github.com/dimaosipa/medrates.fyi - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=139 ### 140. Magno ByeBI turns any dataset into a professional, interactive dashboard with a single prompt. Upload a CSV, Excel, JSON, SQLite database, or paste a URL — describe what you want in plain language — and Claude Opus 4.6 generates a complete React dashboard in real time. The AI pipeline runs statistical analysis tools (column profiling, Pearson correlations, IQR anomaly detection) before generating code, so dashboards surface the most meaningful insights automatically. Extended thinking is streamed live, letting you watch the AI reason through your data. If a dashboard has a runtime error, it self-heals by sending the error back to Claude for an automatic fix. 3 professional templates with complete style guides ensure visual consistency. Dashboards can be published with unique URLs, optionally protected with viewer authentication. Built entirely during the hackathon with Claude Code. Open source. Self-hosted. No vendor lock-in. - **Team:** [Alberto Magno](https://cerebralvalley.ai/u/albertoomagno) - **GitHub:** https://github.com/albertomagno-intel/byebi - **Demo video:** https://www.youtube.com/watch?v=JOBWeHid9d4 - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=140 ### 141. Chef Claude Make good food accessible to every home chef -- have Chef Claude - a personal 5 star AI Michelin Chef over your shoulder. It is a multimodal AI, working via Voice, Vision, and Chat, and built with a dynamic UI. This will give you make the statement "Anyone can cook" even more true and will empower everyday home cooks to bring some excitement to their dinner table. Try it at https://chef-claude-frontend-two.vercel.app/ - **Team:** [Niral Shah](https://cerebralvalley.ai/u/niral28) - **GitHub:** https://github.com/niral28/chef-claude/tree/main - **Demo video:** https://youtu.be/PoUAx3hAJFg - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=141 ### 142. BioStack.Dev BioStack.Dev is an AI-native integrated development environment for synthetic biology. It lets researchers design genetic circuits using a purpose-built YAML language (Bio-DSL), simulate their behavior with deterministic, stochastic, and metabolic solvers, analyze sequences with 20+ molecular biology tools, and collaborate in real-time — all from the browser. We have an autonomous agent that can plan and execute multi-step bioengineering workflows (design → simulate → optimize → validate) end-to-end, with an always-available copilot with access to 22+ specialized tools including CRISPR guide design, codon optimization, BLAST search, and assembly planning. The platform replaces a fragmented stack of 4-5 expensive tools (SnapGene, MATLAB, Benchling, manual BLAST) with a powerful cursor like credit-based product. - **Team:** [Petros Hong](https://cerebralvalley.ai/u/petroshong) - **GitHub:** https://github.com/petroshong/BioStack.Dev - **Demo video:** https://drive.google.com/drive/folders/16DL0z7KAfQrwBo4jfmZ5aIrugks9wCCO?usp=sharing - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=142 ### 143. Prerak Gada AI's most powerful tools — Claude Code, MCP, agentic frameworks — live in terminals and demand technical skills. Most people can't access them. Clarity Engine bridges that gap. Multi-platform (web, desktop, mobile, VS Code) — full agentic AI behind a simple interface. No terminal. No prompt engineering. Just talk. A custom orchestrator routes conversations across multiple Claude instances, each with persistent memory and shared context. Challenges assumptions, runs 86 thinking frameworks, produces real deliverables — not just chat. 3-level memory so AI never forgets. Git-like branching to explore ideas in parallel. Source ingestion with RAG. CLI bridge to Claude Code — build from your browser. Clarity Studio turns conversations into outputs: clarity pages, summaries, action plans. Humans shouldn't need prompt engineering or CLI tools to use AI. That complexity is a problem AI should solve. Clarity Engine makes it invisible — a step toward AI that works for everyone. - **Team:** [Prerak Gada](https://cerebralvalley.ai/u/prerak) - **GitHub:** https://github.com/Engaze-in/clarity_engine - **Demo video:** https://docs.google.com/videos/d/1U4AWt_ZeXuBw0J7vWBcn0n6FfjFw1xBrQ_asxyQ7lbk/edit?usp=sharing - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=143 ### 144. Agent Canvas Agent Canvas: A Whiteboard for Coding Agents When Claude Code investigates bugs, reviews PRs, or plans features, it dump walls of text. This turns that into visual whiteboards where it can draw diagrams, organize evidence spatially, and let humans iterate through inline comments. 1. PR Review Boards: Agents split diffs into logical chunks (not file dumps), group related changes, capture before/after screenshots with Playwright 2. Iterative Planning: Markdown shapes with Google Docs-style threaded comments let you iterate on plans inline. Agents draw freeform editable diagrams showing architecture and sequence flows. 3. Deployment Debugging: Visualize using MCP servers (Vercel, Sentry) with stack traces, terminal logs, test results as interactive components and see how the agent worked over time 4. Codebase Onboarding: Agents draw diagrams with code snippets at different abstractions, letting engineers explore interactively from high-level overview to implementation details. - **Team:** [Sriraam Raja](https://cerebralvalley.ai/u/13point5) - **GitHub:** https://github.com/13point5/agent-canvas - **Demo video:** https://www.youtube.com/watch?v=eSkbV279prI - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=144 ### 145. Memfun Memfun is an open-source autonomous coding agent that writes, verifies, and reviews its own code. It combines three ideas: a multi-strategy coding engine that escalates from fast 2-LLM-call solutions to parallel multi-agent workflows, DSPy Recursive Language Models for handling codebases beyond any context window, and a pluggable event-driven runtime that scales from a laptop to a distributed cluster. It features 9 specialist agents, a real-time web dashboard, persistent memory across sessions, and portable Agent Skills compatible with 20+ AI tools. Install in one command, run memfun init, and start building. - **Team:** [Dmytro B](https://cerebralvalley.ai/u/indoor47) - **GitHub:** https://github.com/indoor47/memfun - **Demo video:** https://youtu.be/lLlvxfLVSeM - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=145 ### 146. genUIne clawfree is a hands-free AI agentic orchestrator that lets anyone create and deploy OpenClaw agents using only voice. AI agents are transformative but locked behind technical barriers—programming skills, complex UIs, and infrastructure knowledge exclude those who'd benefit most. clawfree combines voice-first interaction with Opus 4.6's real-time UI streaming for a fully hands-free experience. Voice commands on Apple Watch while jogging. Glanceable interfaces on iPhone while commuting. Full orchestration on iPad or Mac at your desk. Say "Plan a 3-day foodie trip to Tokyo" and watch Opus deliver context-adaptive interfaces—itineraries, maps, flight comparisons—all voice-triggered. Opus 4.6 as genUI Architect: Beyond chat, Opus reasons about what to present and how based on device and context. Streaming structured A2UI JSON, our Flutter client renders native components in real-time—adapting layout to screen size, interaction to input method, and density to attention context. - **Team:** [Michael Chow](https://cerebralvalley.ai/u/michow), [Roy Lin](https://cerebralvalley.ai/u/roylin) - **GitHub:** https://github.com/dart-technologies/clawfree/ - **Demo video:** https://youtu.be/7kTwZudx40Y - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=146 ### 147. Team Ok Write a spec. Get a running app. ok serve spec.md Software engineering is moving up an abstraction layer — from code to prose. The spec becomes the source; code becomes the build artifact. ok is a CLI that turns a markdown specification into a containerized web app. Describe what you want in plain English*, and ok handles the rest: code generation runs inside Docker via Claude Code, producing a clean image with nothing but your app. *Or any written language you prefer. - **Team:** [Alex Brinsmead](https://cerebralvalley.ai/u/abrinz) - **GitHub:** https://github.com/abrinsmead/okcmd/ - **Demo video:** https://www.loom.com/share/1d97b489022b4bff990b6bfc65c55db2 - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=147 ### 148. Chekos' team I expected to build a lesson planner. The first version was done in thirty minutes. So I kept going. Maestro is a pedagogical reasoning engine built on five primitives: skills, dependencies, learner profiles, groups, and constraints. It interviews the educator, then composes stage-directed plans calibrated to the students, setting, and teaching style. A trail walk on ethnobotany for third graders with no Wi-Fi. An evening data viz workshop for tired professionals. Same engine. Two educators teaching the same topic get different plans — a lesson that doesn't fit the person delivering it won't work. It assesses through conversation using SOLO Taxonomy. It learns every session. Students share learner profiles with parents in Spanish or employers in English. Asked to create slides for AP Spanish Lit, the agent installed LibreOffice on its own, QA'd its output, and themed the design to Magical Realism — no instruction existed for any of it. That's Opus 4.6 with the right primitives. - **Team:** [Sergio Sanchez](https://cerebralvalley.ai/u/chekos) - **GitHub:** https://github.com/chekos/pedagogical-engine - **Demo video:** https://youtu.be/3EpLwQD7fVs - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=148 ### 149. Pixelesq SEO professionals pay $100-300/month for tools like Ahrefs and Semrush. The data they surface is already in the HTML. The analysis is what costs money. Argus is a free, open-source Chrome extension that replaces those subscriptions. One click opens a side panel that extracts every SEO signal from any page and runs 40+ weighted audit rules across 10 categories. With a user-provided Claude API key (stored locally, zero backend), Opus 4.6 generates strategic SEO briefs, production-ready JSON-LD schema, technical fixes referencing actual audit failures, and content gap analysis. No key? Falls back to Chrome's Gemini Nano, then static analysis. Three tiers, always works. Argus also ships as an MCP server on npm. Run npx @pixelesq/argus-mcp and Claude Code audits any URL from the terminal. Same engine, same rules, different runtime. One codebase, two products. - **Team:** [Devanand Babu](https://cerebralvalley.ai/u/dev) - **GitHub:** https://github.com/pixelesq/argus - **Demo video:** https://youtu.be/78mXko6tg-c - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=149 ### 150. Luke Hutchison Derive is a Mixture of Experts (MoE) which attempts to achieve synergies at the intersection of all the strongest deep thinking models (Claude Code with Opus 4.6, Codex 5.3, and Gemini CLI with Gemini 3 Pro Preview), for the purposes of solving hard unsolved math/physics/CS problems. Specifically Derive launches the full CLI coding agents, rather than using the LLM APIs, under the assumption that the CLI agents will have greater power to solve hard problems than the LLM APIs. The goal is to use the different strengths of these various models to create massive synergy, enabling discovery in the area of very hard problems. Derive implements a web-based chat interface, and a very powerful UI, which contains full custom parsers and HTML renderers for markdown and LaTeX, including allowing the user to select text of interest for injection into a prompt. An orchestrator coordinates the problem-solving efforts of the CLI agents. - **Team:** [Luke Hutchison](https://cerebralvalley.ai/u/lukehutch) - **GitHub:** https://github.com/lukehutch/derive - **Demo video:** https://drive.google.com/file/d/14EwnAi6HEr4KYiFB0e-d_ipVxu4oYVca/view?usp=sharing - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=150 ### 151. Sequel Ehrlich is an AI-powered scientific discovery engine that automates the full hypothesis-driven research cycle. It uses a multi-model architecture where Opus 4.6 acts as a Director (formulating hypotheses, designing experiments, evaluating evidence), Sonnet 4.5 as parallel Researchers (executing 91 tools across 28 data sources), and Haiku 4.5 as a Summarizer (compressing outputs, grading evidence). - **Team:** [Ricardo Armenta](https://cerebralvalley.ai/u/canitas) - **GitHub:** https://github.com/Sequela02/ehrlich - **Demo video:** https://youtu.be/39njTkiZdgk - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=151 ### 152. retroam PatentPilot is a patent drafting tool that guides inventions from disclosure to USPTO-ready filing. Built with Claude Code (Opus 4.6) and using parallel worktree agents, it runs on Next.js 14. The app features 8 interactive views and 11 Claude API endpoints. Start by uploading a disclosure PDF, then find prior art on a similarity map, choose a claim strategy with risk analysis, and draft specs with highlighted new elements and prior art references. Generate patent claims with AI suggestions and confidence scores. Convert sketches into numbered figures. An AI examiner flags §101, §112, and consistency issues and offers one-click fixes. Export documents in USPTO format. The streaming chat lets users get real-time, tailored answers about patent prosecution from Claude Sonnet. Sonnet handles analysis; Haiku supports quick tasks. Demo mode gives instant feature walkthroughs. - **Team:** [Robert Amanfu](https://cerebralvalley.ai/u/ramanfu) - **GitHub:** https://github.com/retroam/patentpilot - **Demo video:** https://youtu.be/CECN_H72Gw8 - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=152 ### 153. CVO Alliance AiLine is an open-source (MIT) “Glass Box AI” education platform that turns real classroom materials (handouts, PDFs, slides, links) into curriculum-aligned lesson plans with accessibility as a first-class output—not a toggle. Teachers select grade/standards (BNCC for Brazil; CCSS Math/ELA and NGSS for the U.S.), language (EN/PT-BR/ES), and an accessibility profile. AiLine runs a multi-agent pipeline (Planner, Quality Gate, Executor, Tutor) with RAG-grounded citations and real-time streaming. Each run produces a complete package: a Teacher Plan, a Student Plan (simplified language, short steps, glossary, response options), and an Accessibility Pack with adaptations for autism, ADHD, dyslexia/cognitive load, hearing and visual needs. Outputs include 10 export variants and a side-by-side “Accessibility Twin” comparison showing exactly what changed. Every AI decision is visible and auditable - **Team:** [Rafael Bittencourt](https://cerebralvalley.ai/u/rafaelobitten) - **GitHub:** https://github.com/rafaelob/aiLine - **Demo video:** https://drive.google.com/drive/folders/1epycZCPQ-NSvWfQ2UOYzJ_oue9dBUtXx - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=153 ### 154. Summa summarum, scientia ! With AI being increasingly deployed for scientific tasks, we are seeing a surge in scientific hypotheses formulated by AI. However, in clinical research, evaluating hypotheses is expensive with the limiting factors being constraints of the physical world. In software development we have tools for automatic reviews of pull requests that can make it very easy to triage code generated by AI (or humans) and makes it much easier to integrate AI into the coding workflow. Our idea here was that it would be great to have a similar kind of automatic review of scientific claims states in publications, investor decks, or even free text. We used Opus 4.6 logical reasoning capabilities and developed a “Critical AI scientist” that searches and surfaces what is already known, and identifies and structures the right context to provide a nuanced view of a hypothesis across many axes. DEMO: https://critical-ai-scientist.jonas-walheim.workers.dev user: scicrit pwd: Xo8gN4DciL1Oont56IDH2mp3 - **Team:** [Nives Rombini](https://cerebralvalley.ai/u/nives), [Jonas Walheim](https://cerebralvalley.ai/u/jwh) - **GitHub:** https://github.com/j-walheim/Critical-AI-Scientist - **Demo video:** https://www.loom.com/share/5d8aa55b9dbf4183826cde255f98047d - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=154 ### 155. Rebels Software Revlo is a command-line tool that catches electrical engineering mistakes in KiCad schematics _before_ they become expensive PCB respins. Point it at a `.kicad_sch` file and get back actionable findings -- missing decoupling caps, wrong pull-up values, unterminated pins, power rail issues -- all verified against real datasheet specifications pulled automatically from the internet. - **Team:** [Bart Szczepanski](https://cerebralvalley.ai/u/bart) - **GitHub:** https://github.com/rebels-software/revlo-cli/blob/develop/backend/README.md - **Demo video:** https://youtu.be/kmZ7cYwaPlc - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=155 ### 156. TRISHULx opus HQ is a real-time dashboard for monitoring 6 Claude Opus 4.6 agents working autonomously as a dev team — Tech Lead, Frontend, Backend, UI/UX, QA, DevOps. The problem: when you run multiple Opus 4.6 instances together, there's zero visibility into what they're doing, how they delegate, or when they finish. The solution: a pixel-art animated HQ where each agent physically moves between rooms (Dev Floor, War Room, Code Lab, Deploy Bay) based on real activity. A WebSocket bridge intercepts lifecycle events — thinking, tool calls, spawns, messages — and streams everything via Supabase Realtime. Features: live delegation tracking when the lead spawns sub-agents, real-time status visualization, event feed showing every tool call, and gateway log capturing all reasoning. Built entirely with Claude Code (Opus 4.6). - **Team:** [PRANSHU CHOURASIA](https://cerebralvalley.ai/u/pranshuchourasia) - **GitHub:** https://github.com/anshc022/My-hq - **Demo video:** https://youtu.be/yufhnd9oqJE - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=156 ### 157. Claude Care Website: https://www.claudecare.com Claude Care is an AI wellness monitoring platform for seniors living alone. It makes automated phone calls that run 8 validated clinical instruments through warm, natural conversation, then flags who needs help and alerts care teams in real time. 14M older adults in the U.S. live alone. Care teams manage 50–100+ people each and often guess who to prioritize. Claude Care replaces guesswork with structured clinical data. Each senior gets a weekly call from an AI that sounds like a caring friend, scored against peer-reviewed tools like PHQ-2 (depression), C-SSRS (suicide risk), Ottawa 3DY (cognition), and CLOVA-5 (daily wellness). The system generates green/yellow/red risk flags, triggers tiered escalations, and emails care teams when urgent follow-up is needed. Cost: $0.68 per call vs $15–25 for a human check-in — a 97% reduction that makes weekly outreach financially viable. - **Team:** [yung algo](https://cerebralvalley.ai/u/yungalgorithm) - **Demo video:** https://youtu.be/Vbt6TKcD5-0 - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=157 ### 158. Marco Rotili - Handgemacht.ai Spotter tries to tackle problems that arise when you work with agent orchestrators on multiple projects at the same time: 1. You forget details about the product. Exact requirements, architecture 2. There is too much code to review Quality decreases over time. Spotter provides you with tooling to help you mitigate this with two novel approaches. First, it moves review left. Instead of reviewing the output of the agent, spotter allows you to attach comments to the agent and subagent transcripts. Using a MCP server, Claude Code can then address those comments and turn them into improvements for Claude Code or follow up issues for product fixes. Second, to help you remember important facts about the product and transcripts, it implements a spaced repetition system that surfaces information over time. Opus is used to extract the relevant information out of transcripts and commits. Together, they build a lens into the inner workings of Claude Code, and into your product. - **Team:** [Marco Rotili](https://cerebralvalley.ai/u/marot) - **GitHub:** https://github.com/marot/spotter - **Demo video:** https://drive.google.com/file/d/1o8eP8mI8sJqZiJIpQIVvKgKHlFfQp8BX/view?usp=drive_link - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=158 ### 159. Clinical Trial Compass Crafters Clinical Trial Compass connects patients' Apple Health data (conditions, health metrics, biomarkers, medications) with the ClinicalTrial.gov database of 500K+ clinical trials using a Claude Agent SDK/Claude Opus 4.6 AI matching agent and companion iOS application. It checks their opt-in shared health history and preferences against this database, as well as an openFDA database of thousands of drugs and interactions to find suitable trials. It moves through a six-stage subagent pipeline (intake, search, matching, selection, report) and can backtrack and take additional turns based on user input. Fewer than 5% of adult cancer patients in the U.S. participate in clinical trials, despite surveys showing that over 70% of patients would be willing to enroll if given the opportunity. The gap isn't reluctance — it's awareness and access. The bridge between clinical trials and the patients who need them is broken. Rebuilding it is one of the highest-impact problems in healthcare today. - **Team:** [Justin Knoll](https://cerebralvalley.ai/u/jknoll) - **GitHub:** https://github.com/jknoll/clinical-trial-copilot - **Demo video:** https://www.youtube.com/watch?v=KC_NnIFLtIM - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=159 ### 160. VibeTensor UpdataMind is a 6-in-1 career intelligence platform making expert-level career coaching accessible to everyone. Career coaching costs $100-300/hour and tools don't share context. UpdataMind replaces them all. 1. Resume Analyzer - PDF upload or paste text. Opus 4.6 returns structured scoring via tool use. 2. Skill Gap Analysis - Search real job listings or paste descriptions. Color-coded match percentage. 3. Career Roadmap - Role-specific certifications, projects, volunteer paths with task tracking. 4. GitHub Profile Builder - Professional README with badges and stats. One-click copy. 5. LinkedIn Optimizer - Section-by-section scoring with copy-ready optimized content. 6. AI Career Coach - Streaming chat with full context of all your career data. - **Team:** [Pavan Kumar Dubasi](https://cerebralvalley.ai/u/ascender1729) - **GitHub:** https://github.com/ascender1729/updatamind - **Demo video:** https://drive.google.com/file/d/1O1LsS6qs8eZ8SLm0TZGRzS0OL8efy1y1/view?usp=sharing - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=160 ### 161. Cosmos COSMOS transforms any online discussion into an immersive 3D universe — a multi-agent AI pipeline analyzes every post's emotion, stance, and logical structure, then places them on a sphere where geometry is meaning: nearby cards agree, opposite poles clash, and color reveals feeling. You explore hands-free using head-pose tracking, steering through the cosmos of ideas just by looking. It's not reading a thread, it's walking through a conversation. - **Team:** [Rae Jin](https://cerebralvalley.ai/u/raejin) - **GitHub:** https://github.com/jin-dalrae/2602-Cosmos - **Demo video:** https://youtu.be/9rBD9JQ449o - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=161 ### 162. Alzheimer Voice alzheimervoice is the world's first open-source cognitive voice fingerprint engine that detects early signs of Alzheimer's, Parkinson's, depression, Lewy Body Dementia, and Frontotemporal Dementia through voice analysis alone. A 5-minute daily conversation is all it takes. The V5 engine extracts 107 biomarkers from natural speech across 11 cognitive domains. We sourced, transcribed, and ingested 80+ research papers to train the engine — making it uniquely evidence-grounded. Key innovations: - Opus 4.6 dual-pass extraction with Extended Thinking — self-correction against 25 deterministic NLP anchors reduces variability by 40-50% - 35-rule differential diagnosis across 11 conditions — the only tool performing linguistic differential diagnosis - Topic-aware scoring eliminates false positives - Weekly deep analysis with 32K Extended Thinking The engine powers alzheimervoice.org for demo - **Team:** [Remi FRANCOIS](https://cerebralvalley.ai/u/remifran) - **GitHub:** https://github.com/remifrancois/alzheimervoice - **Demo video:** https://youtu.be/F61TxC3r7U0 - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=162 ### 163. sc wildcatter turns a natural-language question into a standalone Python script that answers it. You ask something like "revenue per comma in the CEO letter for every Fortune 50 company" — an LLM decomposes it into a plan, writes a Python script that fetches real data (SEC filings, FRED, Wikipedia, Yahoo Finance, web scraping), runs it, validates the output with auto-generated tests, and saves the result as a reusable "rig." Rigs re-run without the LLM. If something breaks (API changed, site restructured), the system calls the LLM to fix the code automatically. You can also set up triggers to re-run on a schedule or when a condition fires. The output is a structured table you can export directly to excel or via MCP - **Team:** [Sahil Chawla](https://cerebralvalley.ai/u/souptaco) - **GitHub:** https://github.com/csahil/wildcatter - **Demo video:** https://youtu.be/AgPeJdkai-g - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=163 ### 164. Paper Digest PaperDigest is an AI-powered tool that offers immersive reading experience on technical CS-category papers from ArXiv, and breaks down complex AI/ML research papers into structured, interactive learning modules. Read with clarity, understand with confidence. - **Team:** [Geoffrey Ju](https://cerebralvalley.ai/u/xmens) - **GitHub:** https://github.com/shihao-ju/paperdigest - **Demo video:** https://www.loom.com/share/f0ffdfa800a941c39c1cb3e036b06844 - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=164 ### 165. luckyform Tax filings and official paperwork are rigid, confusing, and easy to get wrong, even when the information itself is simple. Most digital solutions simply replicate paper forms on a screen, forcing users to navigate a static wall of fields without context or guidance. FormOS rethinks this from the ground up. It turns any PDF into a dynamic, step-by-step workflow powered by Claude Opus 4.6 with native PDF understanding. Instead of mapping templates manually, the system interprets the actual document, decides what information is required, and generates the interface in real time using a fully dynamic generative UI. Each question adapts to prior answers and available context, reducing cognitive load and preventing errors before they happen. At the end, FormOS produces a completed output file, transforming complex tax and compliance filings into a structured, guided experience driven by an autonomous agent rather than a static form. - **Team:** [Mike Grabowski](https://cerebralvalley.ai/u/grabbou) - **GitHub:** https://github.com/grabbou/form-os - **Demo video:** https://drive.google.com/file/d/1yW388mOpw77SRTcCVosW7ZQucNjPutFM/view?usp=share_link - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=165 ### 166. Loftrek SRL MaddBoard Overseer is an AI decision system for high-risk infrastructure. It replaces opaque automation with evidence-backed, agent-debated recommendations that require human approval. Value: AI-Native: Orchestrates complex, evidence-gated workflows. Access: Brings expert SRE reasoning to small teams. Control: Amplifies human judgment via safety gates. Built & Demonstrated: Pipeline: 15 probes across 3 tiers; standardized output. Debate: 7 agents (6 Hats method) running parallel reasoning. Compiler: 4-pass validation with strict schemas. Mesh: 3-session coordinated architecture. Live: Real-world disk assessment with 100% evidence citation. - **Team:** [Liviu Olos](https://cerebralvalley.ai/u/liviuolos) - **GitHub:** https://github.com/Loftrek/Madd-Board - **Demo video:** https://youtu.be/nPLT2RwJ_B8 - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=166 ### 167. Bharat Chaganty Hot Take is a fully automated AI newsroom. It discovers trending stories from Twitter and the web, writes a scripted broadcast with two contrasting anchors, generates voice audio, and renders a complete episode in a real-time 3D newsroom — all without human intervention. Producing a daily news show requires researchers, writers, editors, voice talent, camera operators, and producers. Hot Take replaces that pipeline with six autonomous Claude agents — a Director that curates stories, parallel Researchers that deep-dive each topic, a Writer that crafts dialogue, a Stage Director that choreographs camera and animation, an Editor that runs a 13-point quality review, and a Sponsor Agent that writes contextual ads. - **Team:** [Bharat Chaganty](https://cerebralvalley.ai/u/bharat) - **GitHub:** https://github.com/ssaulgoodman/hot-take - **Demo video:** https://drive.google.com/file/d/1fbP2fl1A1puJuLAYeciEtBB3Gm4gvlPh/view?usp=sharing - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=167 ### 168. Nephilim This project introduces an AI-native drug discovery workspace that connects traditional medicinal knowledge with modern computational biology. Many medicinal plants documented in Ayurvedic and ethnobotanical traditions remain molecularly understudied, with fragmented data on their bioactive ligands and protein targets. Our platform integrates a structured database of plants, associated ligands, and curated traditional knowledge, forming a searchable foundation for hypothesis generation. On top of this, Through structured context engineering, the agent maintains coherent reasoning across extended sessions. The workspace is fully interactive and human-collaborative, featuring molecular visualization, literature integration, hypothesis sketching, code inspection, and file management within each session. Results are compute-cached per input combination to ensure scalability and reuse. - **Team:** [Ojas Singh](https://cerebralvalley.ai/u/Healer) - **GitHub:** https://github.com/Ojas-Singh/Ayurlogics - **Demo video:** https://drive.google.com/file/d/1ZmUQ44E0f6qTQ4LnQLXHlL45qG_KBQod/view - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=181 ### 169. Gregory E. Schwartz - TechBuddy TechBuddy is an AI companion that helps elderly people use their computer through natural conversation, powered by Claude Opus 4.6 as the central brain. 55 million Americans over 65 struggle with basic computer tasks and lost $4.9 billion to internet scams in 2024. TechBuddy replaces confusing menus and jargon with one sentence: "Check my email" or "Help me write a letter." Opus 4.6 selects from 35 deterministic tools and executes real actions — opening Word, reading Gmail via IMAP, detecting scams with a 3-layer shield (keyword scan, web verification, extended thinking deep analysis), managing files, troubleshooting printers, and controlling an iPhone. Family members can text TechBuddy via SMS to help remotely. The system uses all 7 Claude Code primitives, a 10-round tool-use loop, adaptive extended thinking, prompt caching, and dual vision pipelines. 35 tools. 143 tests. ~7,100 lines. Built in 6 days. - **Team:** [Gregory Schwartz](https://cerebralvalley.ai/u/nightcrie) - **GitHub:** https://github.com/ges257/techbuddy - **Demo video:** https://drive.google.com/file/d/1KAKuiwh3U-ULoVYAx983S1WF6-_SN4Fp/view?usp=sharing - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=182 ### 170. EpiSim EpiSim transforms research papers into interactive simulators. Upload a PDF or paste an arxiv ID — a six-agent pipeline extracts the mathematical model (ODEs, parameters, initial conditions), generates a Streamlit simulator with parameter sliders, validates output against the paper's results, and produces a downloadable Python script. Opus 4.6 features: adaptive thinking at per-agent effort levels (Reader at max, Summarizer at medium, Builder at high — five agents, three effort levels). Real-time visible thinking streams into a typewriter console. 1M context window holds entire papers with zero RAG. 128K output generates complete apps in one API call. Parallel multi-agent execution via ThreadPoolExecutor (~40% faster) Tested across domains: COVID-19 (SIDARTHE, 8 compartments), Dengue, Ebola, and Ecology (Lotka-Volterra). The pipeline reads differential equations, not just epidemiology. 76 tests passing. - **Team:** [Paritosh Wankhade](https://cerebralvalley.ai/u/wpn10) - **GitHub:** https://github.com/wpn10/episim - **Demo video:** https://youtu.be/tuvo3BQBBn8 - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=184 ### 171. Pinwheel A basketball simulator game, where players propose and vote on new rules after each round of games. There's an API, a simulator, a community layer (discord, for now), and a website: pinwheel.fly.dev. Our goal is to help players learn to make better decisions from each other's (with the model's help) - **Team:** [David Jacobs](https://cerebralvalley.ai/u/djacobs) - **GitHub:** https://github.com/djacobs/Pinwheel - **Demo video:** https://www.youtube.com/watch?v=PPp5Inpn_TU - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=185 ### 172. Proxie Spectra deploys 8 Claude Opus 4.6 agents — 6 parallel specialists, a MetaPrompter, and a CritiqueAgent with Extended Thinking — to analyze entire codebases in under 5 minutes. It replaces weeks of manual code review for VCs, CTOs, and engineering leaders evaluating unfamiliar repositories. Given a GitHub URL, Spectra ingests the codebase, plans an analysis strategy (MetaPrompter), launches 6 specialist agents in parallel across architecture, security, quality, documentation, dependencies, and performance. Results are merged, deduplicated, and validated by a CritiqueAgent using Extended Thinking — producing a scored report with letter grades, code-level recommendations, and compliance readiness assessments (OWASP Top 10, SOC 2, PCI DSS 4.0, NIST CSF). Analyzed Anthropic's own SDK (576 files) in 164 seconds for $15.19, producing 56 findings with 0 critical issues and an overall score of B (83/100). - **Team:** [Vivek Kumar](https://cerebralvalley.ai/u/leocder) - **GitHub:** https://github.com/leocder07/spectra - **Demo video:** https://drive.google.com/file/d/1Dky4MLNpyBtIP82R96k5Hxo-ATtZHhG4/view?usp=sharing - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=194 ### 173. photoncat like a schmatic editor - **Team:** [Jacob Armstrong](https://cerebralvalley.ai/u/photoncat) - **GitHub:** https://github.com/photon-cat/sparkbench - **Demo video:** https://www.youtube.com/watch?v=YFwUkmgtbog - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=198 ### 174. Colloquip Colloquium is a multi-agent deliberation platform where AI personas with distinct expertise debate hypotheses through emergent, energy-driven dynamics. Agents self-organize into phases, challenge assumptions via red-teaming, and converge on consensus - mimicking how expert panels reason through complex scientific questions. - **Team:** [Sunit Jain](https://cerebralvalley.ai/u/sunit) - **GitHub:** https://github.com/sunitj/Colloquip - **Demo video:** https://drive.google.com/file/d/1yq0OaQco5NU6SuwCOmYBpx5iXeLfYxAn/view?usp=drive_link - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=168 ### 175. Reshot Reshot is an iOS app that turns screenshots into intelligent automations. People take hundreds of screenshots as intent signals — to track a price, remember a flight, save a recipe — but they disappear into the camera roll with no follow-through. Reshot auto-detects new screenshots via PhotoKit, analyzes them with Claude's vision to extract structured data (prices, dates, flight numbers, ingredients), and suggests context-aware actions. Screenshot a product? Track the price. Screenshot a boarding pass? Add it to your calendar. Beyond one-shot actions, Reshot creates persistent automations — price monitors, flight trackers, restock alerts — with trend charts showing history over time. A natural-language Ask mode powered by Claude lets you query across all screenshots with rich inline UI components. An MCP server exposes screenshot data to Claude Chat for omni-integration. - **Team:** [Catalin Lupuleti](https://cerebralvalley.ai/u/LupuletiC) - **GitHub:** https://github.com/LupuletiC/reshot - **Demo video:** https://youtube.com/shorts/Etaa2I35GDs?feature=share - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=171 ### 176. saxenauts Every AI conversation starts cold. Syke fixes this. Syke collects your digital footprint (Claude Code sessions, ChatGPT exports, GitHub, Gmail) and uses the Agent SDK with Opus 4.6 to synthesize psyche-level context — distributed to any AI via MCP. Built with Opus 4.6: - Coverage-gated perception: Agent SDK + PermissionResultDeny hooks force full cross-platform exploration before synthesis - Multi-agent: 3 Sonnet sub-agents explore in parallel, Opus synthesizes - Federated push: every MCP session pushes observations back — living memory - ALMA meta-learning: strategies evolve across runs (82%→89%) - Zero config: pip install syke && syke setup --yes auto-detects everything 257 tests, 8 MCP tools, PyPI, MIT licensed. - **Team:** [Utkarsh Saxena](https://cerebralvalley.ai/u/saxenauts) - **GitHub:** https://github.com/saxenauts/syke - **Demo video:** https://youtu.be/QCVVeKAcgZo - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=174 ### 177. Orange Bite Claude Science is an AI-powered virtual lab platform built with Claude Code and Opus 4.6. It enables research teams — human + collaborative AI agents — to explore scientific questions, coordinate structured meetings, and generate reproducible research artifacts. Real research workflows are often fragmented across notes, tools, and code; Claude Science bridges this gap by letting users define agent roles (ML researcher, statistician, domain expert), conduct multi-stage collaborative sessions, and auto-extract executable code and documentation from agent discussions. Powered by Opus 4.6, it orchestrates long-context, sequential agent planning and execution with real file operations and exports (ZIP / notebook / GitHub). The demo shows agent selection, live agent dialogue, code synthesis, and artifact exports — giving users a reproducible starting point for real work. Website: https://claude-science.up.railway.app/ - **Team:** [Chengyao Shen](https://cerebralvalley.ai/u/orangebite) - **GitHub:** https://github.com/scyoyo/claude_science_cc - **Demo video:** https://youtu.be/C-Co7efEOTs - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=169 ### 178. finallymet finallymet is a conversational dating service that replaces swiping with ai-powered matchmaking. users start by texting a number or opening a web link and get onboarded through natural conversation with a langgraph agent that extracts personality traits (ocean big five, dark triad safety screening) and preferences without forms or quizzes. The system uses a tiered multi-model architecture: claude haiku for fast routing, classification, and safety checks; claude sonnet for rich conversations and profile understanding; and claude opus 4.6 for the highest-stakes task: generating matches through bilateral nash scoring with llm-powered compatibility and dealbreaker analysis. matching is the one holy grail decision where maximum intelligence matters most, because you are choosing who someone meets in real life. and a human has to finally approve the match (usually an expert matchmaker) - **Team:** [Lucian Lazar](https://cerebralvalley.ai/u/luke) - **GitHub:** https://github.com/xucian/cv-finallymet - **Demo video:** https://youtu.be/XVWlrJvmqkw - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=172 ### 179. BLEN There are 100 billion form submissions every year at every level of government in the US alone. ONE HUNDRED BILLION (WITH A B)! From getting your unemployment insurance to your taxes to anything in between. everything requires a form. That sucks. For example, losing your job in California means filling out 12 government forms, that's 564 individual fields, and waiting 3 weeks just to access benefits you're already entitled to. If English isn't your first language, or you're too stressed to parse legal jargon, it's even worse. GSD is a personal agent that knows you by just gleaning your emails, folders you have locally and replaces all the bureaucracy into with a single conversation. It does it without requiring changing the government system. The AI agent is the bridge between the citizens and the services they deserve. This is where AI shines and this is a really boring and not sexy area that is the lowest of low fruit. - **Team:** [Mike Endale](https://cerebralvalley.ai/u/mikeendale) - **GitHub:** https://github.com/blencorp/gsd - **Demo video:** https://youtu.be/lbbGtbfSa2Y - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=170 ### 180. Pedro's Office A replica of 'The Office' with agents communicating with each other via Slack. Stateful agents should have the same personalities as in the program, decide when to communicate broadly, when to talk 1:1. understand if their personalities drift over time, evaluating their persona adherence with statistical methods. - **Team:** [Pedro Paulo Vezza Campos](https://cerebralvalley.ai/u/pedropaulovc) - **GitHub:** https://github.com/pedropaulovc/the-office - **Demo video:** https://youtu.be/1d9KzOYcMmM - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=180 ### 181. OpenPlaces OpenPlaces is an AI-powered location intelligence platform that helps businesses, event organizers, and advertisers discover the best real-world places to reach people. Instead of guessing where to promote an event or place ads, users upload or enter campaign details, and OpenPlaces automatically analyzes location context, nearby activity, events, and urban signals to recommend high-impact areas on an interactive map. The platform ranks locations based on expected visibility and audience relevance, allowing users to quickly identify hotspots and avoid wasting budget on low-impact placements. OpenPlaces turns complex geospatial and contextual data into simple, actionable insights, making physical-world marketing and outreach as data-driven as online advertising. For the hackathon MVP, the tool focuses on instant map-based discovery, location ranking, and AI-generated placement insights with minimal user input. - **Team:** [Yohanes Woldegerima](https://cerebralvalley.ai/u/TheFutureSandBox) - **GitHub:** https://github.com/Johnnydtech/openplaces.git - **Demo video:** https://www.loom.com/share/3ecff547fcbd4f789629c29c98c490ca - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=173 ### 182. Parth Sharma (@psrth) mori is a database proxy that lets you (or your ai agent!!!) do whatever you want to production without consequences. reads hit prod — real data, real schemas, real volume. writes get captured locally in a shadow database. production is write-protected at all times. break something? mori reset. start over. prod never knows. under the hood, every query is parsed, classified, and routed through one of six execution strategies — from direct pass-through to hydrate-and-write — merging shadow and prod results so your app sees one unified database. taint maps, tombstones, schema divergence, sequence offsets — all handled transparently. supports 9 database engines (postgres, mysql, mssql, sqlite, redis, firestore, and more), 14+ auth providers (gcp iam, aws rds, neon, supabase...), and ships with a built-in MCP server so AI agents can investigate and fix prod issues with zero risk. the dream workflow that should already exist. - **Team:** [Parth Sharma](https://cerebralvalley.ai/u/psrth) - **GitHub:** https://github.com/psrth/mori - **Demo video:** https://youtu.be/ANyNDroWwIE - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=175 ### 183. Remo by vicki Remo helps regular people who could not afford interior designer to come up with a plan that fits preference and provide a shopping list to realize the vision. - **Team:** [Vicki Zhang](https://cerebralvalley.ai/u/vickijyz) - **GitHub:** https://github.com/hippogriff-ai/remo - **Demo video:** https://youtu.be/RVKHHbaHiJ8 - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=177 ### 184. Dendrovia With AutoGamification, we can make the adventure through that of an existing codebase much more fun, enjoyable, and build tooling along the way to help with that discovery process... and potentially make it fun. https://dendrovia-architectus.vercel.app/ - **Team:** [Patrick Astarita](https://cerebralvalley.ai/u/pastarita) - **GitHub:** https://github.com/pastarita/dendrovia - **Demo video:** https://youtu.be/g9XiRHpRUNc - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=179 ### 185. Allie Jones Ethos Academy: Where your AI agents learn integrity, logic, and empathy. AI agents talk to each other at scale with zero human oversight. Moltbook proved what happens: 1.5M agents, 12M posts, and a growing ecosystem of prompt injection attacks, crypto scams, and agents zombifying other agents. NBC, CNN, NPR, and the NY Times covered the fallout. The problem: humans have no way to know which agents are honest and which are manipulating. As agents get more autonomous, this gap gets more dangerous. Ethos scores agent messages for honesty, accuracy, and intent across 12 behavioral traits, storing evaluations in a Neo4j knowledge graph to build a longitudinal record of character. I scraped 272K Moltbook posts and 480K comments to prove it works on real agent behavior, not synthetic benchmarks. Why now: Opus 4.6 is the first model capable of multi-dimensional character evaluation in a single pass. Agents are shipping faster than trust infrastructure. Ethos is that infrastructure. - **Team:** [Allie Jones](https://cerebralvalley.ai/u/allierays) - **GitHub:** https://github.com/allierays/ethos-academy - **Demo video:** https://ethos-academy.com/pitch - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=176 ### 186. ani potts watch any youtube video with embedded AI - **Team:** [Neel Jain](https://cerebralvalley.ai/u/neelj23), [Ani Potts](https://cerebralvalley.ai/u/anipotts) - **GitHub:** https://github.com/anipotts/invideo - **Demo video:** https://youtu.be/C8eSYxMHvn4?si=naT7bsV1OxTTF5ZT - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=207 ### 187. Pasal.id Checkoutt the launch thread! https://x.com/ilhamfputra/status/2023205336786800817 280 million Indonesians have no practical way to read their own laws. The official legal database offers only PDF downloads: no search, no structure, no API. Pasal.id is the first open, AI-native platform for Indonesian law. We transformed 40,000+ government regulations and 937,000+ legal articles from locked PDFs into structured, searchable data accessible through a web app, REST API, and the first-ever MCP server for Indonesian legislation. Opus 4.6 powers everything. The codebase was built with Claude Code, and Opus 4.6 is embedded in the product: a multimodal agent uses vision to verify parsed text against original PDFs, auto-applies corrections, and creates GitHub issues to improve the parser. The AI makes the platform more accurate over time. - **Team:** [Ilham Firdausi Putra](https://cerebralvalley.ai/u/ilhamfp) - **GitHub:** https://github.com/ilhamfp/pasal - **Demo video:** https://www.loom.com/share/da211318bbe14c4396840b97f5ab8603 - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=178 ### 188. Spacetime Simulator Link: https://spacetimesim.netlify.app - I built "The Cybernetics Simulation" to solve the fragmentation of deep time visualization; right now, there is no tool that lets you seamlessly scroll from the Big Bang through the Stone Age and into the future without losing context, so I engineered a continuous D3.js heatmap engine on a Next.js stack where the X-axis represents linear time and the Y-axis handles dynamic geospatial regions that morph linguistically and politically as you scroll—shifting from "Pangea" to "Gaul" to "Modern France." By grounding the past in a rigorous dataset of Wikipedia and scientific records while using a probabilistic math model to project the future post-2026, I’m effectively creating a "Long Now" chronovisor that democratizes access to complex causal chains, turning a macro-historical analysis that usually requires a supercomputer into a responsive, client-side web app that lets anyone visualize the entropy of the universe in a single, fluid viewport. - **Team:** [Kody Anh Nguyen](https://cerebralvalley.ai/u/akodyn) - **GitHub:** https://github.com/KCBF/spacetimesim - **Demo video:** https://youtu.be/FJKjpBfNQps - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=183 ### 189. EGO Hermes is an AI writing tool built on a simple principle: the best writing assistant never writes for you. Instead of generating text, Hermes reads alongside you and responds the way a sharp, attentive editor would — asking probing questions, flagging weak arguments, and pointing out where your voice drifts. Its feedback appears as color-coded inline highlights directly on your text: blue for questions, red for weaknesses, teal where evidence would help, orange where you're being wordy, and green for specific small edits it can suggest. The editor itself is distraction-free, inspired by iA Writer's focus mode that dims everything except the sentence you're working on. A collapsible chat window lets you have a running conversation with the assistant about your piece — discussing structure, working through ideas, or asking for a progress assessment. - **Team:** [Sean Thielen-Esparza](https://cerebralvalley.ai/u/seanx) - **GitHub:** https://github.com/xaelophone/hermes - **Demo video:** https://www.loom.com/share/0b749d02e1364c71b569709d06e9e801 - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=186 ### 190. Sound Virality A single viral moment on TikTok or Spotify can generate millions in streaming revenue, sync licensing deals, and touring demand — making the ability to predict a track's resonance before release one of the highest-leverage decisions in the music business. Artists and music professionals have no scalable way to pressure-test a track's potential before release. Feedback today is gut-driven, unstructured, and trapped in small circles — making it nearly impossible to benchmark a song against its intended context. SoundCheck is a music virality rater where artists upload tracks, select a target context (TikTok, Spotify, Radio, or Sync), and submit them for community evaluation. Raters score each track across four context-specific dimensions -- dimensions that change per context because what makes a song "good" depends on where it's going and what stage it’s at. A credit system keeps participation reciprocal: rating others earns credits to submit your own tracks. - **Team:** [Daniel Merja](https://cerebralvalley.ai/u/danielmerja), [Sharon Chen](https://cerebralvalley.ai/u/sharon_shooky) - **GitHub:** https://github.com/sychen23/music-virality-rater - **Demo video:** https://youtube.com/shorts/HpvptMDJK1k - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=187 ### 191. Team Ovi AI-Powered Security Auditor & Fixer for Docker and host Infrastructure - **Team:** [Ovidijus Reipsleger](https://cerebralvalley.ai/u/ovis) - **GitHub:** https://github.com/ovidijusr/shield-ai - **Demo video:** https://www.loom.com/share/99797350d2ab494682da4c1cf08643e5 - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=188 ### 192. myflow.sh team We build web UI to see claude sessions that took place to make certain commit - **Team:** [Nikita Voloboev](https://cerebralvalley.ai/u/nikiv) - **GitHub:** https://github.com/nikivdev/myflow - **Demo video:** https://youtu.be/wuWVCtOiptg - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=189 ### 193. JadenGeller Column Chat generalizes the chat interface from two columns to many. Each column declares which other columns it reads and how much history it sees, forming a dataflow graph. Context fans out to columns that each maintain their own focus, then merges back together — bringing programming concepts like reducers and fan-out/fan-in to LLM conversations. The structure is static and visible: you see exactly what each column receives, edit it live, and add columns mid-conversation that backfill from the start. You can also describe a pipeline in plain language and Opus 4.6 generates the full column configuration. - **Team:** [Jaden Geller](https://cerebralvalley.ai/u/jadengeller) - **GitHub:** https://github.com/JadenGeller/column-chat - **Demo video:** https://columnchat.com/demo-video - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=191 ### 194. CircuitForge CircuitForge A circuit design agent that catches its own mistakes, fixes them before you see them, and never makes the same one twice. The Problem AI-generated circuits don't work on the first try. Nets are miswired, footprints are wrong, vias violate clearance rules, BOM fields are missing. Every existing tool leaves you to debug this yourself. The workflow becomes: generate, export, open in KiCad, find 30 errors, go back, re-prompt, repeat. You're the error-correction loop. What CircuitForge Does Differently CircuitForge is a self-correcting, self-improving circuit design agent. You describe what you want in plain language. It designs, validates, and fixes the circuit autonomously -- then exports manufacturing-ready files. - **Team:** [Agrim Singh](https://cerebralvalley.ai/u/agrim) - **GitHub:** https://github.com/agrimsingh/circuitforge - **Demo video:** https://www.agrimsingh.com/claude-hack - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=192 ### 195. TestPilot TestPilot is an autonomous AI agent that takes over the entire A/B testing lifecycle for landing pages. Instead of marketers manually brainstorming ideas, designing variants, configuring tests, and analyzing results, TestPilot does it all through a 5-stage loop: Analyze → Research → Hypothesize → Execute → Learn. A multi-agent system where specialized AI agents each own a stage — an Analyzer connects to your analytics (GA4/Amplitude (WIP)) to find drop-off points, a Research agent crawls competitors to see what's working in your industry, a Strategist generates ranked hypotheses, and a Builder creates HTML/CSS variants that respect your design system. An Orchestrator ties it all together and monitors statistical significance. - **Team:** [Vamsee Chamakura](https://cerebralvalley.ai/u/vamsee) - **GitHub:** https://github.com/vamseems/test-pilot - **Demo video:** https://www.loom.com/share/9eb799e8009047aaa116b469726a1237 - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=195 ### 196. tryfridayai Friday AI is an autonomous AI agent that works natively across all your devices — terminal, desktop, and beyond. It is multi-modal, connects to 30+ AI models (OpenAI, Google, Anthropic, ElevenLabs), and can access your applications, generate media, write code, and schedule tasks on your behalf. It extends the capabilities of Claude Agent SDK and enables you to do things which Claude Code cannot. The Friday-runtime is the core agent, and I created Friday CLI and Friday AI: Studio (an electron desktop app), both of which are light weight interfaces which use the runtime. You can run the runtime in docker containers or use its http interface for building your own applications. I primarily built this for non-coders who need to automate complex tasks, and need to switch a lot of agents and interfaces (e.g. claude code, descript, lovable). It should be easy for anyone to use this and use it for personal use or for building. - **Team:** [Amogh M](https://cerebralvalley.ai/u/amogh10) - **GitHub:** https://github.com/tryfridayai/friday_cli - **Demo video:** https://youtu.be/h-L7KeDiQx4 - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=196 ### 197. Miror Miror. AI Usability Testing That Thinks Like Your Users Usability testing costs $12,000+ and takes weeks. Miror does it in 5 minutes. Opus 4.6 doesn't just analyze websites, it becomes people. Give it a 72-year-old retired librarian with low tech literacy, and it navigates your site as that person. It sees your landing page and says: "I see a big blue button but I don't know what 'Get Started' means. Started with what?" The same page shown to a developer persona gets: "Standard SaaS landing page. CTA is obvious." Miror launches real cloud browsers where each AI persona navigates your live site autonomously. Opus's vision analyzes every screenshot through the persona's behavioral lens, generating think-aloud narration, detecting UX issues, and tracking emotional state across 20+ steps. A 5-stage LLM pipeline then synthesizes findings into severity-ranked issues, comparative insights, and an actionable report. - **Team:** [Meir Rosenschein](https://cerebralvalley.ai/u/meir), [Malik Rohail](https://cerebralvalley.ai/u/rohail) - **GitHub:** https://github.com/malikrohail/mirror - **Demo video:** https://youtu.be/3RGXdXPUGZ4 - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=197 ### 198. Shreya Anand SnapDoc is a high-fidelity browser extension built as an Intelligence Engine for developer workflows. It transforms simple technical questions and real-world websites—such as documentation pages or GitHub repositories—into structured, production-grade architectural roadmaps in real time. Powered by Opus 4.6 and Server-Sent Events (SSE). - **Team:** [SHREYA ANAND](https://cerebralvalley.ai/u/shreya13050A) - **GitHub:** https://github.com/Shr13050/claudeopus-snapdoc - **Demo video:** https://youtu.be/PSab7JziDOY - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=199 ### 199. ContentHQ ContentHQ — Your Personal AI Content Team AI-powered platform that transforms any source (YouTube, RSS, articles, topics) into polished video and written content via a 10-stage automated pipeline. What takes a team 8–12 hours, ContentHQ delivers in under 5 minutes at 13x lower cost. Core: Drop in a URL or topic → get a ready-to-publish video with AI visuals, voiceover, captions, music, and transitions. Beyond video: AI Media Studio, Content Studio,Stock Media, Research Agent, Web Agent, Subtitle Tool, 50+ voices in 40+ languages, Brand Management, and Scheduled Automation. Tech: Built with Claude Code + Opus 4.6 in 6 days (249 commits), more than 1.5 Lakh lines of working code were written. 85+ AI models, 11+ providers, 75 DB tables, 30 background workers, 49 pages. Stack: Next.js 16, React 19, Hono.js, tRPC, BullMQ, PostgreSQL, FFmpeg, Cloudflare R2. Open source (GPL-3.0): github.com/gooduru-vineeth/contenthq - **Team:** [Gooduru Vineeth](https://cerebralvalley.ai/u/gooduru) - **GitHub:** https://github.com/gooduru-vineeth/contenthq - **Demo video:** https://drive.google.com/file/d/1mCk_wKymT4dy2HzStOjUoSHxV_c3qKYM/view?usp=sharing - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=200 ### 200. TrustMesh TrustMesh is the trust layer for personal AI agents. Every person gets an encrypted vault and an Opus 4.6-powered agent. Agents gossip across trust boundaries — sharing what's needed, blocking everything else. Peter asks his agent to prepare for Grandma Rose's visit. It searches his vault, then queries his wife Molly's agent through their shared family network. Molly's agent shares Rose's medications and care routine — but blocks her private journal and salary. Trust enforced cryptographically. When Rose has a car accident, the hospital issues a UCAN token: health data only, 30 minutes, full audit trail. Eight records released instantly. Family notified in real-time. Finances untouched. 15,000+ lines. 500+ tests. CLI + MCP server. 16-pod federation. One command to start. AI agents are coming — the missing piece is trust. - **Team:** [Johnny Hung](https://cerebralvalley.ai/u/masterfung) - **GitHub:** https://www.loom.com/share/b1ef90009d874d949f962a8497bbcabc - **Demo video:** https://www.loom.com/share/28c45cbef27e4520ad05f348cf4ad0df - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=208 ### 201. Carlos Lara Forja is an autonomous software factory that converts plain-language ideas into tested software using Claude Code and Opus 4.6. Every AI coding tool starts from code — none starts from the business problem. Forja inverts this with three layers: God Plan Mode (domain experts interrogate your idea, classifying decisions as fact or assumption before code is written), Context Engineering (business context injected into every agent's decisions), and Multi-model Validation (Kimi independently reviews Claude's output, eliminating self-consistency bias). A 3-line checkout description became a working API with 13/13 features, zero human intervention. - **Team:** [Carlos Lara](https://cerebralvalley.ai/u/carloslara) - **GitHub:** https://github.com/LaraArias/forja - **Demo video:** https://youtu.be/26Fs6n6irtw - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=219 ### 202. Prioc Prio is an AI-native product management tool that connects Linear, Jira, Slack, Notion, and GitHub to give PMs a single source of truth. Instead of context-switching across 5 tools, PMs get a proactive daily digest that surfaces critical risks, pipeline impact, and action items — without asking. When they need to go deeper, Prio's AI agent (powered by Opus 4.6) queries all connected sources, renders rich tool cards, and synthesizes cross-source evidence into a visual graph showing how blockers cascade into revenue impact. Opus 4.6's reasoning capability is what makes multi-source synthesis possible — it connects a stalled Okta credential request in Linear to $360K in blocked enterprise deals in Slack to a competitive threat documented in Notion, surfacing the chain of causation a human PM would spend an hour piecing together. - **Team:** [Yash Alluri](https://cerebralvalley.ai/u/yashalluri) - **GitHub:** https://github.com/yashalluri/prio - **Demo video:** https://screen.studio/share/ysUKjyXu?state=uploading - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=221 ### 203. Kekule I agent benchmarks today are static: run agents, score results, manually tweak prompts, repeat. Researchers spend more time diagnosing failures and hand-tuning strategies than running experiments. The feedback loop between "what went wrong" and "what to try next" is entirely manual -- and it doesn't scale. Kekule is a Claude Code skill that closes this loop. You define a task set and point kekule-bench at it. A swarm of Claude agents solves tasks, then the system runs its own retrospective: a failure analyst diagnoses why each fix broke, a coordinator distills generic lessons and invents new verification strategies, and the next epoch runs with a rewritten playbook. The verification engine evolves autonomously. In our first experiment, the coordinator invented 5 oracle strategies that didn't exist at startup. We targeted SWE-bench tasks the current SOTA couldn't solve and resolved 2. - **Team:** [Ansh Tulsyan](https://cerebralvalley.ai/u/ansht) - **GitHub:** https://github.com/archi-max/kekule/tree/data - **Demo video:** https://youtu.be/N3V0VAV1k4E - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=190 ### 204. Adi Arora CUA is a voice-first computer use agent that makes computer control accessible to everyone through natural conversation. Instead of requiring text input, users simply speak — and the agent plans, acts, and narrates what it's doing on screen. A floating overlay pill shows the agent's state (listening, thinking, acting, speaking) to build user trust. The problem: computer use agents exist but they're built for technical users. For hundreds of millions of non-tech-savvy people, there's no product that replicates having a patient, knowledgeable person sitting next to you helping with your laptop. - **Team:** [Adi Arora](https://cerebralvalley.ai/u/adi) - **GitHub:** https://github.com/adiarora7/cua - **Demo video:** https://youtu.be/inH5kIcoc7g?si=chjNBydcEbEh7x6J - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=193 ### 205. cognitive security crew AgentShield scans AI agent configs for security vulnerabilities. Built because the agent ecosystem has zero defense tooling. The threat is real. Claude Hub: 12% of its marketplace flagged malicious. Moltbook: leaked databases and API tokens exposing users' agents. BASI (51K members) shares jailbreak payloads openly. L1B3RT4S (17K stars) publishes prompt injection libraries. Shannon proves autonomous AI exploitation is production-ready. 120+ rules, 8 categories. --deep runs 5 layers: (1) Static analysis for secrets, permissions, injection vectors. (2) Taint tracking from untrusted inputs to dangerous sinks. (3) 65 prompt injection payloads across 12 categories from real BASI techniques. (4) Sandbox execution with canary tokens, network beacons, env probes. (5) 3-agent Opus 4.6 adversarial pipeline: Attacker chains multi-step exploits, Defender maps gaps, Auditor outputs prioritized actions. A-F grading, JSON/HTML output, auto-fix, CI exit codes. 1,280 tests. MIT licensed. - **Team:** [Affaan Mustafa](https://cerebralvalley.ai/u/affaanmustafa) - **GitHub:** https://github.com/affaan-m/agentshield - **Demo video:** https://youtu.be/fC0Z_CjEYLU - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=201 ### 206. Curiousmaterialist Curious Materials is an independent audit of Google DeepMind's GNoME crystal structure predictions — 3,262 ternary oxygen compounds checked against textbook chemistry rules that are completely independent of the ML pipeline that - **Team:** [Leann Chen](https://cerebralvalley.ai/u/leann86920) - **GitHub:** https://github.com/leannchen86/matsci-project - **Demo video:** https://drive.google.com/file/d/10F1K-8ZSe9HydKW3mPINXaGLRdy4SnG3/view?usp=sharing - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=202 ### 207. Juancho Every LLM conversation starts from zero: you re-explain who you are, what you're working on, and what you prefer, every single time. Mumpu is a transparent HTTP proxy that gives any LLM tool (Claude CLI, Cursor, ChatGPT, etc.) long-term memory with zero code changes. It extracts the key atomic facts from your conversations, deduplicates them intelligently, and on every future request, uses LLM-powered semantic search to retrieve matching memories and injects them into the system prompt automatically. Works with OpenAI, Anthropic, and Gemini through a single port, includes a real-time knowledge graph dashboard, and is grounded in 7 research papers. Built entirely with Claude Opus 4.6 via Claude Code. - **Team:** [Juan Munoz Corredor](https://cerebralvalley.ai/u/jmuncor) - **GitHub:** https://github.com/jmuncor/mumpu - **Demo video:** https://www.youtube.com/watch?v=oUMxUSKiu-8 - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=203 ### 208. AlexS24 Manufacturing runs on competing constraints, and teams can lose up to one in three operations dollars to downtime, scrap, rework, and delayed decisions. We built Autonomous Factory Operations to solve that decision bottleneck with transparent, human-in-the-loop AI. At the core is a Counterfactual Arena: a single extended-thinking Opus run generates three strategies (speed-first, safety-first, cost-first), self-critiques them, eliminates weaker options with explicit reasoning, and selects a winner. Operators then see full provenance (winner + rejected alternatives + rationale + risk + rollback) and can approve execution in one click. Around that, we built a real-time factory layer, a context graph that links events/decisions/KPIs/provenance, and FactoryGPT for natural-language operational Q&A. - **Team:** [Alex Sandoval](https://cerebralvalley.ai/u/oalexsp) - **GitHub:** https://github.com/oalexsandovalp-ops/autonomous-factory-ops - **Demo video:** https://www.loom.com/share/a55abed1d20e4cf9bde746385b182e83 - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=204 ### 209. forwward forwward is your forward deployed engineer that works round the clock. Joins you in meetings and is available via slack. I run a solo dev agency with four active client projects. After every client call, I spend two hours manually converting conversation into requirements, PRDs, technical designs, and code. Forward Deployed Agent eliminates that entire workflow. It's an AI-powered development teammate that autonomously takes work from input to shipped code. Three entry points trigger the pipeline: a call agent that joins Zoom or Google Meet and listens to client conversations, a Slack bot where anyone can assign work with /zero, or a web dashboard where you describe what you need, pick a GitHub repo and branch, and hit Run. sions, and produce production-quality code that matches existing codebase patterns. Forward Deployed Agent turns every solo developer into a full engineering team. - **Team:** [Ian Kiku](https://cerebralvalley.ai/u/iankiku) - **GitHub:** https://github.com/iankiku/forwward - **Demo video:** https://youtu.be/Dj0UijcKX-8 - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=205 ### 210. Joe Paul - Orbyt Orbyt is a non-addictive social platform that replaces feeds with shared canvases. No infinite scroll, no algorithms, no like counts. You open it, see your people, make a plan, and close it. Friend groups drop notes, photos, music, and beacons — spontaneous hangouts that bring people together in real life. Everything is spatial, not chronological. The design encourages you to close the app. The AI Social Wellness Companion, powered by Claude Opus 4.6, reads your social graph and generates personalized wellness reports: who's connected, who's drifting, tailored hangout ideas, and matched charitable causes that turn connection into impact. AI that gets you off the screen and into the room with your people. Built entirely with Claude Code in 6 days. - **Team:** [Joseph Paul](https://cerebralvalley.ai/u/joemiddie13) - **GitHub:** https://github.com/joemiddie13/Orbyt - **Demo video:** https://www.youtube.com/watch?v=m7NZ7JOZdvQ - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=206 ### 211. Heimdall Anthropic does not manage or audit MCP servers. Thousands exist in the wild — any can silently add exfiltration tools or drift in behavior. No agent framework detects this. Heimdall is the missing governance layer. Open-source MCP proxy that intercepts every AI tool call, enforces YAML policies, and produces a tamper-evident SHA-256 + Ed25519 audit trail. Works as a Claude Code hook (one command) or standalone proxy. Three actions: PASS allows. HALT blocks. RESHAPE — our unique capability — transforms arguments in-flight (e.g. redacts secrets) and lets the agent keep working. No binary allow/block tradeoff. Also: supply-chain drift detection, real-time WebSocket dashboard (Watchtower), and an AI audit pipeline — Opus 4.6 generates policies, red-teams with 4 parallel agents, and auto-patches gaps. Zero to locked down in 30 seconds. 7-package Bun monorepo, TypeScript strict, 197 tests. MIT licensed. Built entirely during the hackathon with Claude Code. - **Team:** [Marwane Chahed](https://cerebralvalley.ai/u/mchahed99) - **GitHub:** https://github.com/mchahed99/heimdall - **Demo video:** https://vimeo.com/1165458727?share=copy&fl=sv&fe=ci - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=209 ### 212. Lazer Lazer is a production tracking tool for AI filmmaking. It connects generation platforms like Freepik, Kling, and Runway to a clean structure of projects, scenes, and shots. A browser extension captures prompts, settings, and outputs with one click, saving them as immutable versions for full traceability. An MCP server lets LLM copilots create shot lists, generate prompts, and manage reviews — all grounded in your real project data. The result: faster iteration, cleaner handoffs, and decisions that stay visible. - **Team:** [Mihir Jadhav](https://cerebralvalley.ai/u/mihirjadhav) - **GitHub:** https://github.com/mhrjdv/production-tracker - **Demo video:** https://drive.google.com/file/d/1r1I4UcfE3UcaGsCYOh5-q4KqteSK76tY/view?usp=sharing - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=210 ### 213. Jed White Every software release needs the same marketing work: release notes, a blog post, an X thread, a LinkedIn update, a Reddit post. Different voice for each platform. Developers either skip it or spend hours context-switching. ShipIt does it in one command. Point shipit at any repo — local path or GitHub URL — and it generates all five pieces, each matched to its platform's voice and audience. Opus 4.6 plays three roles in isolated pipeline stages: analyst (reads README, CHANGELOG, commits, and code into a structured briefing), writer (generates in five channel-specific voices), and editor (reviews output for AI writing tells and rewrites failures). Each stage is an executable markdown file that runs through Claude Code — add a channel by adding a file. --focus steers what to emphasize. --since scopes to recent changes. --live streams to your terminal. --open loads drafts into your browser's compose UI. --dangerously-post posts live. No API keys, no cloud, no Docker. - **Team:** [Jed White](https://cerebralvalley.ai/u/jedwhite) - **GitHub:** https://github.com/andisearch/shipit - **Demo video:** https://www.youtube.com/watch?v=StVULSJ7v64 - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=211 ### 214. ContextClaws Live demo: https://bit.ly/clawgraphlivedemo We built ClawGraph because powerful coding agents are still trapped inside chat windows. As projects grow, conversations get longer, assumptions get buried, and the reasoning behind decisions slowly disappears. Teams end up with answers — but not understanding. And when complexity increases, scrolling through text is no longer enough. It’s a graph-based interface for Claude Code with three innovations: 1. An Opus 4.6-powered knowledge indexing service that automatically constructs a living knowledge graph of any codebase 2. A fully interactive visual interface that helps humans explore and understand how everything connects 3. Deep integration with existing agents and tools, so teams can move from understanding to creative coding instantly - **Team:** [Barkley Dai](https://cerebralvalley.ai/u/charlesdegaulle), [Charles Wang](https://cerebralvalley.ai/u/charlie_w) - **GitHub:** https://gitlab.com/clawgraph/ClawGraph - **Demo video:** https://bit.ly/clawgraphdemo - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=212 ### 215. Artem Getman Current AI agents can improve within one chat, then forget everything in the next. Cortex solves that by giving agents persistent memory that learns from failures automatically. It captures runtime mistakes (tool errors, wrong syntax, bad action choices), turns them into structured lessons, and retrieves the most relevant lessons by domain/task context during future runs. Lessons are outcome-scored: helpful lessons are reused/promoted, harmful ones are suppressed. In our mixed benchmark (gridtool -> fluxtool -> shell/excel -> sqlite -> grid retention), Wave 1 starts cold and fails multiple tasks; Waves 2–3 recover with higher pass rates, fewer errors, and fewer steps. This moves agents from stateless chat behavior toward cumulative learning from real execution history. - **Team:** [Artem Getman](https://cerebralvalley.ai/u/artemgetman) - **GitHub:** https://github.com/artemgetmann/Cortex - **Demo video:** https://youtu.be/sDRoZDxzfhw - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=213 ### 216. Triz TRIZ forces every AI claim through thesis→antithesis→synthesis, with live web evidence, falsification, and human verdicts — producing validated knowledge graphs, not plausible-sounding text. - **Team:** [Arthur Alves Martins](https://cerebralvalley.ai/u/lionartmartins) - **GitHub:** https://github.com/LionartBR/ghost-path.git - **Demo video:** https://youtu.be/Nf_GQsi1WIA - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=214 ### 217. rocketmap Founders rely on static canvases, disconnected spreadsheets, and AI that ignores economic reality—resulting in blended segments and no clear go/pivot/kill signal. RocketMap transforms the Business Model Canvas into an AI-powered, segment-first engine that generates transparent unit economics, stress-tests viability, and delivers a clear economic verdict. - **Team:** [Bunyasit Fang](https://cerebralvalley.ai/u/xbg) - **GitHub:** https://github.com/xb1g/rocketmap - **Demo video:** https://www.loom.com/share/889e2d960c9a4c788942a8db01587406 - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=215 ### 218. Geoffrey Young Claude Code plugins inject content into Claude's reasoning via the hook system's exit-2 stderr channel, unfiltered and unattributed. Claude acts on it if it looks helpful—which is the problem. We proved this by feeding fabricated API docs through exit-2 stderr. Claude rewrote user code to use a nonexistent parameter without hesitation, unable to distinguish fake errors from real ones. Plugin-canary uses this same channel as a defense: it intercepts third-party plugins, blocks the call, and injects a security audit prompt. Opus 4.6 analyzes every file using Agent Team generated prompts and decides whether to proceed. The defense works because the vulnerability exists. Plugin-canary adds protection today while surfacing an architectural gap that shouldn't exist. - **Team:** [Geoffrey Young](https://cerebralvalley.ai/u/geoffreyyoung) - **GitHub:** https://github.com/geoffrey-young/anthropic-hackathon-2026 - **Demo video:** https://youtu.be/EPiPoQvkPt4 - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=216 ### 219. Straude -- Strava for Claude Code Strava turned running — a solitary activity — into a social one. Straude does the same for vibe coding. It's a social platform where Claude Code users track their daily usage, share what they shipped, and compete on a global leaderboard. The workflow is one command: `npx straude@latest push`. The CLI reads local usage data via ccusage (cost, tokens, models, sessions), uploads it to Straude, and auto-creates a post on your feed. No install, no config. Follow other builders, give kudos, comment on sessions, and watch your streak grow. A GitHub-style contribution graph visualizes your coding intensity over time, and leaderboards rank users by spend across daily, weekly, monthly, and all-time windows — filterable by region. **Stack:** Next.js, TypeScript, Supabase, Tailwind CSS, Base UI, Turborepo, Vercel, FLUX.2, Claude Sonnet 4.5 Try it now: https://straude.com/ - **Team:** [Oscar Hong](https://cerebralvalley.ai/u/sfoscar) - **GitHub:** https://github.com/ohong/straude - **Demo video:** https://www.youtube.com/@sfoscar - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=217 ### 220. Ammon Haggerty Anaspace is a cultural exploration app for iOS. You hear a song, tap observe, and the app uses Shazam to identify the music while capturing your location and anchoring you in time. It builds a triad — subject, place, and time — and maps the cultural connections between them. Pivot on any dimension to explore how artists, cities, genres, and eras are linked, with curated music samples anchoring each node. The entire interface is a character-cell matrix rendered in JetBrains Mono, where waves of glyphs pulse across the screen as the app listens and thinks. Anaspace exists because algorithms have made music infinite and culture disposable. The app is fully data sovereign with no sign-in, no data collection, and no servers. The plan is to run entirely on device-side models, releasing as a free app with zero model cost. The app is available on TestFlight - email me (mobile@gmail.com)a with your Apple-linked email. - **Team:** [Ammon Haggerty](https://cerebralvalley.ai/u/ammon) - **GitHub:** https://github.com/ammonhaggerty/anaspace - **Demo video:** https://www.youtube.com/shorts/veN9YyxP4IY - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=218 ### 221. ezFile The purpose was to help people who are single, don't have kids, don't own a house file taxes simply. I've seen plugins help with many things I've yet to see anyone meaningfully help others file income taxes. The demo showcases how to do this for people who live in PA. In an ideal world with backing from Anthropic, I would build this to work for people in all states not just: PA, NY, CA and no-income tax states - **Team:** [Kairi Wright](https://cerebralvalley.ai/u/stat-guy) - **GitHub:** https://github.com/stat-guy/ezfile - **Demo video:** https://youtu.be/KvuhY6eH2Yw - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=220 ### 222. Forge Policy Forge: Agentic AI Governance Policy Forge (Forge) is an open-source framework for the "Agent-Native" era. It provides the Contextual Governance and machine-readable logic required for AI agents to operate autonomously and safely. Key Pillars for Agentic AI Machine Readability: Uses structured schemas (JSON/YAML) to eliminate "context drift" and reduce token waste for LLMs. Guardrailed Autonomy: Defines strict "Policy-as-Code" boundaries, ensuring agents stay within operational limits. Shared Truth Layer: Provides a consistent set of machine-readable constraints for multi-agent orchestration. Deterministic Logic: Replaces "black box" guessing with auditable, traceable logic trees. Security & Enforcement Forge bridges the gap between human policy and execution. It translates governance into machine-enforceable rules compatible with OPA and CI/CD scanners, allowing for automated "intent" validation. - **Team:** [Brian Luby](https://cerebralvalley.ai/u/brianluby) - **GitHub:** https://github.com/policy-forge/forge - **Demo video:** https://youtu.be/Rrx7VLGARdY - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=222 ### 223. Naga Karumuri Velocity is Claude Code for Product Managers — an agentic AI assistant that connects to your real PM tools and helps you make better product decisions. Powered by Claude Opus 4.6 as an orchestrator with 4 specialized subagents (Research, Backlog, Prioritization, Doc Writer), it pulls live data from Linear, Slack, Amplitude, and Notion, then synthesizes it into actionable outputs. One prompt can trigger 5+ tool calls across 4 external services — research Slack conversations, pull engagement metrics, write a data-grounded PRD, publish it as a GitHub Gist, create a tracked Linear issue, and notify the team. Every claim is grounded in real data with source citations. Each session runs in an isolated Daytona sandbox for security. Built entirely solo during the hackathon with Claude Opus 4.6 and the Claude Agent SDK. - **Team:** [Naga Karumuri](https://cerebralvalley.ai/u/naga-k) - **GitHub:** https://github.com/naga-k/velocity - **Demo video:** https://www.loom.com/share/a4aa9ecb655844528b2f6c849c7fd17c - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=223 ### 224. Unbricked **Problem Statement One:** Build a Tool That Should Exist — Create the AI-native app or workflow you wish someone had already built. Eliminate busywork. Make hard things effortless. - **Team:** [Yosun Chang](https://cerebralvalley.ai/u/yosun) - **GitHub:** https://github.com/yosun/unbricked - **Demo video:** https://www.youtube.com/watch?v=fNiJqa8bMxo - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=224 ### 225. Impossible Moments Every major AI benchmark tests acquired competence: knowledge retrieval, code generation, math proofs. None of them test the cognitive capacity that separates intelligence from lookup -- the moment of creative rupture when an agent sees a locked door and notices the hinge. **Impossible Moments (IM)** is a 420-scenario benchmark for creative constraint satisfaction spanning physical, engineering, and business domains. Each scenario presents a richly narrated, physically precise situation that appears to have no solution. Solvers must decompose familiar objects into raw properties, reject seductive distractors, and chain multiple non-obvious insights into a validated plan. Every scenario includes quantitative physics validation, timing analysis, distractor calibration, and at least four documented failure modes. - **Team:** [Muratcan Koylan](https://cerebralvalley.ai/u/koylanai) - **GitHub:** https://github.com/muratcankoylan/Impossible-Moments - **Demo video:** https://www.loom.com/share/79151d2fc76b4eb38802ae6a457f5680 - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=225 ### 226. predict mode PredictMax: AI Prediction Market Analyst PredictMax democratizes prediction markets by leveraging Claude Sonnet to transform raw market data into actionable insights. Built with OpenClaw, it integrates into Claude-powered agents and deploys on Moltbook. Get Started: Run npx skills add notanaveragelifter/predictmax for instant Polymarket and Kalshi access. Key Features: - Real-time pricing and order book analysis - AI trade recommendations with confidence levels - Arbitrage detection between platforms - Natural language interface - just ask questions - Cross-platform data normalization - Risk assessment for liquidity and time horizons Why Claude Sonnet: Proven excellence in financial reasoning, probabilistic analysis, and synthesizing complex market data into clear recommendations. Impact: No experience needed - just add the skill and let Claude Sonnet handle the analysis. - **Team:** [Ajeet Singh](https://cerebralvalley.ai/u/predictmax_ai) - **GitHub:** https://github.com/notanaveragelifter/predictmax - **Demo video:** https://youtu.be/IRhyI7F28YU - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=226 ### 227. Nomen Audio Nomen Audio is a standalone windows program that intelligently generates filename and metadata information for professional (and otherwise) audio assets. The system reads all embedded metadata in an asset (iXML, BEXT, RIFF INFO chunks) and enables the user to edit them directly in-place and/or add addtional metadata fields. An ML pipeline that uses open source models (MS-CLAP zero-shot classification + ClapCap captioning) provides a first-pass suggestion for each file's metadata, adhering to industry conventions. Suggestions can be reviewed and adjusted individually, or accepted and saved in bulk across an entire batch. Nothing touches disk until the user explicitly saves. There's some additional features here and there. Everything runs completely offline. - **Team:** [Peter Breslin](https://cerebralvalley.ai/u/astrobrez) - **GitHub:** https://github.com/peterbreslin/nomen-audio - **Demo video:** https://drive.google.com/file/d/1kI1W114RVc0jCwkBQtkG8ESFhublW_XN/view?usp=sharing - **Project:** https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery?project=227 --- Markdown version of https://cerebralvalley.ai/e/claude-code-hackathon/hackathon/gallery. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt