# National Security Hackathon (by Army xTech): Project Gallery - **Event:** [National Security Hackathon (by Army xTech)](https://cerebralvalley.ai/e/3rd-annual-natsec-hackathon) - **When:** May 2 at 9:00 AM – May 3 at 4:00 PM (PDT) - **Where:** Shack15, San Francisco, CA - **Hosts:** [Shield Capital](https://cerebralvalley.ai/u/Shield-Capital), [Cerebral Valley](https://cerebralvalley.ai/u/cv) - **Projects:** 102 (5 placed) - **Page:** https://cerebralvalley.ai/e/3rd-annual-natsec-hackathon/hackathon/gallery ## Projects ### 1. Vanguard Vanguard — Real-Time Persistent Surveillance for Palantir Foundry The problem with military surveillance today is that every option requires too much: drones need operators and FAA clearance, fixed cameras need infrastructure, radio reporting takes hands and attention, and nothing writes automatically into Palantir Foundry. Vanguard turns a $300 pair of Ray-Ban glasses into a Palantir-native intelligence sensor. A soldier puts them on and starts walking. The system detects every person in view, assigns them a persistent track ID that survives even when they briefly leave frame, and writes live TrackedPerson objects directly into Palantir Foundry — automatically, in real time, with no operator input. When someone crosses a predefined zone, a ZoneTransition event fires in Foundry. Bounding boxes and track IDs overlay the glasses stream. The soldier's phone runs everything on-device. Nothing requires a tablet, a drone, fixed infrastructure, or someone staring at a screen. The pipeline runs on Meta DAT SDK 0.6.0, Apple Vision for person detection, OC-SORT for persistent tracking, a Bun/TypeScript relay server, and Palantir Foundry Ontology API for live intelligence writes. A web Command Center shows all live streams, per-workflow telemetry, and per-node firing analytics in real time. The US Military already runs on Palantir. Vanguard didn't build another dashboard. It built the sensor that feeds it — hiding in plain sight. - **Team:** [Parnell Thrower III](https://cerebralvalley.ai/u/PThrowerIII), [Elijah Arbee](https://cerebralvalley.ai/u/ebowwa) - **GitHub:** https://github.com/ebowwa/meta-wearables-dat-ios - **Demo video:** https://youtu.be/RHOalwk9Ifk - **Project:** https://cerebralvalley.ai/e/3rd-annual-natsec-hackathon/hackathon/gallery?project=28 ### 2. Shadow Fleet Detector Shadow Fleet Detector is a multi-intelligence maritime monitoring app that fuses Sentinel-1 SAR imagery with AIS vessel data to visualize vessel activity, reporting gaps, and dark vessels on an interactive map. - **Team:** [Darijan Ducic](https://cerebralvalley.ai/u/darijan), [Alessandro Duico](https://cerebralvalley.ai/u/2ico) - **GitHub:** https://github.com/2ico/natsec3-shadow-fleet - **Demo video:** https://www.loom.com/share/260e85692c2941339a15bf59b520fada - **Project:** https://cerebralvalley.ai/e/3rd-annual-natsec-hackathon/hackathon/gallery?project=73 ### 3. SkillsHarness Here's your project description, ready to paste into the submission form: --- SkillHarness: In-Flight Capability Composition for Autonomous Drone Swarms SkillHarness is a microVM skill composition runtime that enables a single operator to dynamically re-task an autonomous drone swarm in natural language, with real FAA airspace enforcement, running entirely on a Palantir-provided Jetson Orin Nano edge device no cloud dependency. Problem: Autonomous systems today are monoliths. Every capability is hardcoded at firmware level. When the mission changes or a new threat emerges, operators must land, reflash, and redeploy — a cycle that takes hours and may not be operationally viable. Integration alone consumes 60-80% of prime contractor budgets. Solution: SkillHarness introduces SKILL.md — a portable capability packaging format analogous to Docker containers for autonomous systems. Skill VMs load, execute, and can be hot-swapped mid-mission via KVM hardware isolation and vSock communication. A navigation skill can be replaced with a jamming detection skill in under 2 seconds, airborne, without rebooting. What we built and demonstrated live: A 3-drone autonomous swarm running on a single Palantir Jetson Orin Nano with 8 hardware-isolated skill VMs. The operator issues natural language FRAGO orders that are parsed, geo-fence validated against real FAA UDDS airspace data (378,000+ features), and distributed to the swarm via Raft consensus over NATS JetStream. When the swarm leader is killed mid-mission, Raft elects a new leader in under 2 seconds and the mission continues. Three live sensors feed a real-time common operating picture: a VFAN u-blox 7 GPS receiver providing actual position with 10 satellites, a Logitech C270 webcam running OpenCV target detection publishing STANAG 4607 ISR contacts, and a NooElec NESDR SMArt v5 SDR receiving live ADS-B aircraft transponders at 1090MHz from the SFO approach corridor overhead. Problem statements addressed: PS2 (Edge Deployments and Drone Operation) primary — we hit all three PS2 examples directly. PS1 (Sensor Analysis and Integration) secondary — real GPS, camera, and RF sensor fusion into one tactical COP. Why it's novel: No existing system supports in-flight capability hot-swap without rebooting. The SKILL.md format creates a portable capability standard that works across any autonomous platform. The combination of Raft consensus, vSock-isolated microVMs, real FAA airspace enforcement, and multi-modal sensor fusion on 15W of edge compute is not available in any commercial or open-source autonomous systems platform today. Beyond military: Emergency response (wildfire drone re-tasking), infrastructure inspection (single platform, multiple inspection skill sets), autonomous vehicles (OTA skill updates without downtime), and offshore energy (ROV capability loading at depth). - **Team:** [Suvasis Mukherjee](https://cerebralvalley.ai/u/suvasis) - **GitHub:** https://github.com/dronomyio/Hackathon_SkillsHarness.git - **Demo video:** https://youtu.be/V2VdBZahTn8 - **Project:** https://cerebralvalley.ai/e/3rd-annual-natsec-hackathon/hackathon/gallery?project=29 ### 4. Eaglemaxxing The intelligence system that can draw connections across an infinite number of data sources, for ANY data type. - **Team:** [Nicole Ge](https://cerebralvalley.ai/u/Nic0le) - **GitHub:** https://github.com/nicole-learn/EAGLEMAXXING - **Demo video:** https://youtu.be/NiPGtWxvuqY - **Project:** https://cerebralvalley.ai/e/3rd-annual-natsec-hackathon/hackathon/gallery?project=74 ### 5. Eagle Robotics A multi-modal CUAS detector that fused RGB, thermal, and event streams. - **Team:** [Craig Iaboni](https://cerebralvalley.ai/u/Craig-Iaboni) - **GitHub:** https://github.com/EagleRobo/eaglerobotics-natsec2026/tree/main - **Demo video:** https://drive.google.com/file/d/10brfYt7fK2XKGzXFBBLWM1zKSKC4PFQ_/view?usp=sharing - **Project:** https://cerebralvalley.ai/e/3rd-annual-natsec-hackathon/hackathon/gallery?project=85 ### 6. FireFly Intel Firefly Intel is on-device tactical sensor fusion for the dismounted soldier. It runs entirely on the tablet the operator already carries, fusing EO, IR, RADAR, and RF feeds into a single threat picture in red, yellow, and green. The problem: a soldier in the field cannot trust single-modality sensors because modern adversaries deploy multispectral decoys that fool thermal and camera sensors, while the real attack comes from another vector. He also cannot send raw video back to base; bandwidth is too small, and latency is too long. He has to decide alone, in seconds. Firefly catches decoys through cross-modal disagreement: when EO and IR confirm a target but RF is silent where it should be loud, the system flags a likely spoof. A quantized Phi-3-mini language model runs locally for plain-English drone tasking, and the same quantized model runs on the drone, so the tablet and drone communicate in kilobytes of metadata, not gigabytes of raw video. Cursor-on-Target XML to TAK. No cloud, no uplink, sub-second decisions. Prototyped under the SFARS program with the U.S. Army Engineer Research and Development Center (ERDC). Built through Hacking 4 Defense at Florida Atlantic University. - **Team:** [Tarruck Wheeler](https://cerebralvalley.ai/u/tarruck) - **GitHub:** https://github.com/Tarrucks/Firefly-MOSAIC - **Demo video:** https://youtu.be/HcTaYjPouTo - **Project:** https://cerebralvalley.ai/e/3rd-annual-natsec-hackathon/hackathon/gallery?project=30 ### 7. Beacon Beacon is an AI-powered, offline-capable system that converts a soldier's voice into doctrinally-correct military reports (MEDEVAC 9-lines, LOGSTATs, contact reports) — no internet required. - **Team:** [Angel Zambrano](https://cerebralvalley.ai/u/angelzzz23), [Kalin Ovtcharov](https://cerebralvalley.ai/u/Kalin) - **GitHub:** https://github.com/kovtcharov/beacon - **Demo video:** https://youtube.com/shorts/NT4orrtrUJo - **Project:** https://cerebralvalley.ai/e/3rd-annual-natsec-hackathon/hackathon/gallery?project=79 ### 8. SAURON We see through walls. Project Sauron is a passive WiFi-based through-wall localization system that detects and tracks human presence inside a structure in real time — without emitting any signal. By reading perturbations in existing ambient WiFi channel state information across three disposable ESP32 nodes, it classifies personnel location by zone in under 100ms. Full system cost: under $50. Undetectable to any EM sensor. https://drive.google.com/drive/folders/1w7ZMn4W8OX5Cau4EOP4xct74m3X6B0-8?usp=sharing - **Team:** [Mario Sumali](https://cerebralvalley.ai/u/mario), [Devin F](https://cerebralvalley.ai/u/foong), [Arnold Yang](https://cerebralvalley.ai/u/Arnold), [Kyleen Liao](https://cerebralvalley.ai/u/kyleenliao) - **GitHub:** https://github.com/kyleenliao/sauron - **Demo video:** https://youtu.be/9Lrj8A5WQ8M - **Project:** https://cerebralvalley.ai/e/3rd-annual-natsec-hackathon/hackathon/gallery?project=88 ### 9. The Covenant 24/7 Maritime Situational Awareness Solution with All-in-one Unmanned System - **Team:** [Seyeon Lee](https://cerebralvalley.ai/u/TheCovenant), [Jacob Nyhagen](https://cerebralvalley.ai/u/JNyhagen) - **GitHub:** https://github.com/SeyeonLee/black-sheep-wall.git - **Demo video:** https://youtu.be/a_c72a7_3Tc - **Project:** https://cerebralvalley.ai/e/3rd-annual-natsec-hackathon/hackathon/gallery?project=2 ### 10. SpectralEye SpectralEye is an electronic warfare system that allows you to represent with high fidelity your current electromagnetic state. SpectralEye uses advanced RF modeling techniques to understand and show the operators how their battlefield looks from the electromagnetic spectrum. It also allows them to create simulations to pen test their electronic warfare setups. SpectralEye is terrain agnostic and is also built on the NVIDIA Siena framework, which is used in cutting-edge research both in electronic warfare systems and 5G and 6G systems, SpectralEye can be dual-use as it can help both municipal powers and the military, but also telecom providers, to understand their electromagnetic spectrum better. - **Team:** [Anton Njavro](https://cerebralvalley.ai/u/njavro) - **GitHub:** https://github.com/njavro/SpectralEye - **Demo video:** https://youtu.be/aVyJxi3Iv64 - **Project:** https://cerebralvalley.ai/e/3rd-annual-natsec-hackathon/hackathon/gallery?project=31 ### 11. UncWorks Hamilton — Trust-Aware C2 Layer for Contested Airspace Hamilton is a real-time command-and-control layer that gives commanders the most relevant comms-integrity information at a glance, so they can issue the correct orders and equip their assets with the right tools to guarantee mission success. We introduce a trust score: a Rust listener ingests Cursor-on-Target telemetry over UDP and post-processes every frame — dedupe gaps, sequence breaks, corruption, jitter, neighbor agreement — to abstract a single value that denotes how compromised a ground unit's comms and signals are. That score feeds a munitions fingerprint system that detects jamming patterns in the degraded telemetry and matches them against likely jamming methods being imposed on the combat assets — so commanders see not just that a unit is compromised, but how, and can issue countermeasures accordingly. - **Team:** [Kristian Romero](https://cerebralvalley.ai/u/kristiansromero), [Joey Osborne](https://cerebralvalley.ai/u/josborne), [evan ly](https://cerebralvalley.ai/u/chheang) - **GitHub:** https://github.com/chheangg/uncwork/ - **Demo video:** https://youtu.be/BE-5zfrQkS0 - **Project:** https://cerebralvalley.ai/e/3rd-annual-natsec-hackathon/hackathon/gallery?project=50 ### 12. AltiAir - DarkMesh Darkmesh is a decentralized tactical command-and-control mesh that survives DDIL environments — Denied, Degraded, Intermittent, Limited connectivity. The DoD's nightmare is a near-peer fight where adversaries jam Starlink and sever cloud links. Most C2 stacks die the moment connectivity drops. Darkmesh keeps the squad's operating picture intact even when the network and the coordinating node both go down. Three Pi-class compute nodes form a closed mesh on a private LAN. Each runs local LLM inference — Gemma 4 E2B on the Jetson Orin Nano, Gemma 3 on the Pi 4Bs — fuses multimodal sensor input from camera, microphone, and RFID, and gossips a shared world state. One node is elected coordinator via Raft-style heartbeat consensus. When the leader is killed mid-pitch, another assumes command in under one second. No data loss, no awareness gap. Darkmesh answers Palantir's hackathon brief directly: orchestrating intelligence across two tiers when the link between them is unreliable. Evidence queues locally, uploads to Palantir Foundry via OSDK when uplink returns, and receives ontology-enriched cues back through the mesh. AIP Logic auto-classifies severity and drafts SITREPs. Danti provides geospatial enrichment when connected. The use case is counter-UAS cueing in controlled training environments. Camera, microphone, RFID, and provider-style location feeds correlate into evidence packages. Operators see policy-gated review cues, never autonomous engagement. Human-in-the-loop on every actionable output. Built in 24 hours by Team Altiair: Sarah Hatcher, Benjamin Shyong, Rob Grossman, Katherine Lambert. Aligned with DARPA's DICE program for distributed C2 resilience, the Pentagon's Replicator initiative for attritable autonomous systems, and USMC MAGTF-scale operations. Around $400 of edge hardware. Zero cloud assumptions. Site: darkmesh.us - **Team:** [Benjamin Shyong](https://cerebralvalley.ai/u/BenjaminBear), [Katherine Lambert](https://cerebralvalley.ai/u/StellaLuminosity), [Sarah Hatcher](https://cerebralvalley.ai/u/skripke), [Rob Grossman](https://cerebralvalley.ai/u/digitalnomd) - **GitHub:** https://github.com/digitalnomd/altiair - **Demo video:** https://youtu.be/rrvOQ09IXpI - **Project:** https://cerebralvalley.ai/e/3rd-annual-natsec-hackathon/hackathon/gallery?project=62 ### 13. BlueMark BACKGROUND - Modern war is a drone war: Ukraine fields ~7M cheap FPV drones in 2026 — now the #1 cause of battlefield casualties, ahead of artillery. - The US is unprepared: No mass-produced $400 drone, no doctrine for swarm-on-swarm attrition — a critical gap if Taiwan, Korea, or the Baltics ignite. - ~50% of friendly drones are killed by friendly fire: No working IFF (Identification-Friend-or-Foe) for low-end drones; EW interceptors jam anything emitting video. - Legacy IFF doesn't fit: Transponders cost more than the drone, are too heavy, and emit RF that lets the enemy target friendlies. - Solving this doubles the arsenal: +3.5M effective drones/yr for Ukraine today; a ready-to-deploy IFF layer for the US before peer conflict. - Ground-truthed with operators: Built with the founder of DWARF Engineering, Ukraine's leading UAV-stack scaleup — kill-zone reality, not Pentagon theory. TECHNICAL - Steganographic watermark, invisible to the enemy: HMAC-SHA256 marker hidden in unused VBI lines of the analog video stream — undetectable without the key, so the enemy can't use it to target friendlies (the failure mode of every other IFF). - Unforgeable and unreplayable: Rolling counter blocks replay; pre-shared key never goes over air; hierarchical keys (master → unit → drone) limit blast radius if a drone is captured. - Zero added BOM, zero new procurement: Firmware-only on the drone (Betaflight + MAX7456 OSD chip); ~$50 receiver on the EW side ($15 capture card + $35 Raspberry Pi). One-frame decision latency (~16ms). - 7-signal fusion, 6 states, one-way safety invariant: Degraded signals can only lower confidence, never raise. Crypto decides; gated YOLOv8n only corroborates — unvalidated ML can never promote a drone to "friendly." - Distribution flywheel: Free Betaflight/Kordyceps integration → MoD-mandated logistics-stage flash (NATO IFF model) → per-drone OEM license → sub-cent pay-per-contact at scale. - Working demo today: ESP32 firmware ↔ Python receiver byte-identical; React mission control; 17/17 tests green. - **Team:** [Sebastian Steen](https://cerebralvalley.ai/u/Seban), [Birger Steen](https://cerebralvalley.ai/u/Birger), [Nicholas Fleischhauer](https://cerebralvalley.ai/u/unogi), [Arpit Mathur](https://cerebralvalley.ai/u/arpitmathur) - **GitHub:** https://github.com/sebaleks/fpv-friendly-id-demo - **Demo video:** https://youtu.be/yVcUMnm2TkI - **Project:** https://cerebralvalley.ai/e/3rd-annual-natsec-hackathon/hackathon/gallery?project=67 ### 14. Hawk-I We built a real-time National Security Digital Twin on Palantir Foundry. Hawk-I integrates live Shodan cyber-telemetry, Exa AI OSINT, and Danti searches into a unified Ontology. By mapping relationships between Hostile_Threats, Logistics_Vessels, and Combat_Units, the system identifies how digital gray-zone attacks (e.g., energy grid compromise) directly degrade physical mission readiness. Features a RAG-powered Commander Chat (Hawk-I) for autonomous Course of Action (COA) generation and a "Blackbox" lineage view for Human-in-the-Loop verification. Commander Agent takes initiative when human-in-the-loop is not, ensuring autonomous, and rapid responses to threats to Supply Chains. - **Team:** [Sathvik Vempati](https://cerebralvalley.ai/u/Honec) - **GitHub:** https://github.com/SVstudent/Hawk-I/tree/main - **Demo video:** https://www.loom.com/share/a706e7cba1704c6bb48db9819ce4bcd7 - **Project:** https://cerebralvalley.ai/e/3rd-annual-natsec-hackathon/hackathon/gallery?project=3 ### 15. Field Viewers We built a Field Viewers–branded web dashboard plus a Raspberry Pi HTTP gateway that turns low-cost FV-13 Scout WiFi drones into a carry-and-fly recon swarm: operators stage with a laptop or MacBook, a Pi, and drones that fit a backpack-scale kit—no cloud data center required for the demo workflows we show. The dashboard unifies mission planning on a satellite-style map, swarm formations and spacing, waypoint routing with return-to-home, live and simulated downlink video, and scenario-driven rehearsal for container inspection, plant perimeter, and medical isotope leak–style missions. A radiation-style dose field is visualized on the ground from per-drone bottom-sensor cones so teams see where signal is hot, not only where aircraft icons sit. A second workflow demonstrates in-browser facial recognition on live gimbal MJPEG from the Pi gateway: an enrolled reference photo is compared to the stream with an on-screen match verdict, useful for identity checks without uploading video to a third party. The project is open-source and public-safe: no credentials, captures, or APKs in the repository; local configuration uses gitignored env files. Tech stack includes a Vite React operator UI, optional mission API, Node gateway on the Pi mirroring XR872-style paths, and documented runbooks for Mac lab, Pi sync, and demo recording. - **Team:** [Sachin Junnarkar](https://cerebralvalley.ai/u/FieldViewers) - **GitHub:** https://github.com/fieldviewer/XTechFieldViewersPublicDemo - **Demo video:** https://youtu.be/cfRiehgtISg?si=CRC7BudoiC4mrYPN - **Project:** https://cerebralvalley.ai/e/3rd-annual-natsec-hackathon/hackathon/gallery?project=4 ### 16. Spencer Brown Open claw to fly a drone. Just the start of an MIT fully open source project. (Could only attend 1/2 hackathon due to a funeral) - **Team:** [Spencer Brown](https://cerebralvalley.ai/u/Spencerbrown1717) - **GitHub:** https://github.com/SpencerBrown1717/Free_Motion - **Demo video:** https://freemotion.tech/ - **Project:** https://cerebralvalley.ai/e/3rd-annual-natsec-hackathon/hackathon/gallery?project=32 ### 17. Gallatin Radio Agent, deployable in the cloud or at the edge on the battlefield. Listens to all available channels, transcribes, categorizes and acts, fully autonomously where appropriate and with human in the loop where needed. Built and tested for both green gear and more advanced MANET radios. Integrated to action outcomes across the Army Maven Smart System and Navigator, a leading tactical resupply software. - **Placement:** 1st Place - **Team:** [Daniel Buchmueller](https://cerebralvalley.ai/u/danbuci), [Justin Oh](https://cerebralvalley.ai/u/justinoh), [Jake Siegel](https://cerebralvalley.ai/u/Jsiegel43) - **GitHub:** https://github.com/gallatinai/xtech26/ - **Demo video:** https://youtu.be/QkbkF2aKfVI - **Project:** https://cerebralvalley.ai/e/3rd-annual-natsec-hackathon/hackathon/gallery?project=45 ### 18. Red Team AI Multi-agent AI platform that tests applications for vulnerabilities. - **Team:** [Illia Oleksiuk](https://cerebralvalley.ai/u/ioleksiuk) - **GitHub:** https://github.com/ioleksiuk/nationalsecurityhack - **Demo video:** https://www.loom.com/share/309b980c352a4b69811f7abe7b8c5a88 - **Project:** https://cerebralvalley.ai/e/3rd-annual-natsec-hackathon/hackathon/gallery?project=47 ### 19. Narya This project is an AI-assisted search-and-rescue sensing platform that fuses LiDAR, thermal, and visible camera data to detect human presence, localize it in 3D, and stream that information to remote operators. In a rescue like the stranded airman mission in Iran, a ruggedized version mounted on a drone, ground robot, or temporary overwatch sensor could have reduced the search area faster, confirmed whether a heat/body signature was actually a person, and given rescuers a live 3D map of likely human locations, obstacles, and approach paths without immediately exposing personnel. It runs fully on the edge and connects directly to the Foundry to integrate the data into Palantir's post-processing pipeline. - **Team:** [Alejandro Pinto](https://cerebralvalley.ai/u/aipinto), [Jonathan Duron](https://cerebralvalley.ai/u/J_d), [Arthur Garzon](https://cerebralvalley.ai/u/arthurgarzon), [Tom Almog](https://cerebralvalley.ai/u/TomAlmog) - **GitHub:** https://github.com/Summma/pids - **Demo video:** https://youtu.be/3igBDjksKW8 - **Project:** https://cerebralvalley.ai/e/3rd-annual-natsec-hackathon/hackathon/gallery?project=49 ### 20. Team Mycelium Mycelium solves the critical bandwidth bottleneck in military Mobile Ad Hoc Networks (MANETs). The military already operates at the state of the art with deeply advanced technologies. Currently, pushing massive operational files (like 10GB 3D targeting maps) to a drone swarm and auditing those nodes for data integrity congests and crashes fragile radio networks. To resolve this, we are building a decentralized tactical data-sharing protocol. It utilizes a peer-to-peer file chunking system to distribute massive files concurrently across edge devices, distributing the bandwidth load. To secure the swarm against compromised nodes without transmitting heavy data back to the commander for auditing, the protocol integrates lightweight, recursive Zero-Knowledge Proofs (Plonky2 zk-SNARKs using Poseidon hashing). Drones recursively verify their neighbors' file chunks and mathematically roll the proofs together. The commander ultimately receives a single ~300-byte proof that guarantees the data integrity of the entire swarm, dropping the bandwidth cost of a security audit to near zero - **Team:** [Dijesh Shrestha](https://cerebralvalley.ai/u/DJ14), [Sarad Gaihre](https://cerebralvalley.ai/u/croxx), [Shishir Poudel](https://cerebralvalley.ai/u/sisirpdl) - **GitHub:** https://github.com/codeitdijesh/XTechTLazarus - **Demo video:** https://youtu.be/2InFL8IcunU - **Project:** https://cerebralvalley.ai/e/3rd-annual-natsec-hackathon/hackathon/gallery?project=56 ### 21. DPI Research In this day and age of National Security, launch cadence is important because we literally want to own the space above earth. The current process to send payload to space is fraught with unnecessary friction. I present to you the Satellite mission planner, a system that will make that process smoother. - **Team:** [Daniel Pang](https://cerebralvalley.ai/u/danp) - **GitHub:** https://github.com/dpiresearch/SatConOps - **Demo video:** https://youtu.be/M44EA7DbwtM - **Project:** https://cerebralvalley.ai/e/3rd-annual-natsec-hackathon/hackathon/gallery?project=5 ### 22. Aora We trained a model imitating rat's danger detection senses into what humans can feel and used bee hive examples to then pass it to other people at the field once early signs of danger are detected. It can then result in better results mission wise as bombs or drones coming from afar that are not visible for human eye can be sensed with our body and that information can then spread to the whole crew. Milliseconds during such missions matter and we managed to get to 500ms earlier detection of danger, with time this algorithm can evolve to be minutes earlier. - **Team:** [Dara Tumenbayeva](https://cerebralvalley.ai/u/dara) - **GitHub:** https://github.com/dara-aora/aora-defense-hackathon-threat-detection - **Demo video:** https://youtube.com/shorts/IhYlEhIYPPc - **Project:** https://cerebralvalley.ai/e/3rd-annual-natsec-hackathon/hackathon/gallery?project=33 ### 23. Sabotage as a Service (SaaS) Autonomous drones sit on 2 extremes, ISR and kinetic effects. We aimed to explore the large gap in between those 2 capabilities. We built a drone capable of surgical sabotage (placing payload on enemy assets) and environment manipulation (opening control panels) through a robotic arm, all on-device. We build an autonomous drone with a robotic arm and fully on-board compute (Jetson) running a quantized LLM acting as the reasoning layer, allowing the drone to autonomously generate mission plans and execute both in drone navigation and dexterous manipulation. The ai brain takes natural language commands and converts them to low-level control primitives and analyses camera feeds for object detection and navigation. - **Team:** [Lucas Souza](https://cerebralvalley.ai/u/lsouza), [Stephen Lacy](https://cerebralvalley.ai/u/stephenlacy) - **GitHub:** https://github.com/Rhodrium/SaaS - **Demo video:** https://youtu.be/BSA4jFluVm0 - **Project:** https://cerebralvalley.ai/e/3rd-annual-natsec-hackathon/hackathon/gallery?project=36 ### 24. The Long Island Gulls Army SignalSentry is an AI-assisted command-and-control platform for disaster response operations. It fuses distress signals from multiple sources into a unified geospatial operating picture, automatically recommends optimal unit assignments using routing, flood-risk, and capability factors, and pushes live updates to all operators via a real-time shared map — replacing fragmented radio logs and spreadsheets with a single tool that helps incident commanders make faster, better-informed decisions under pressure. - **Team:** [Zack Schwartz](https://cerebralvalley.ai/u/zhs), [Gil Valdes](https://cerebralvalley.ai/u/gilvaldes) - **GitHub:** https://github.com/RaythaHQ/natsec-hackathon - **Demo video:** https://youtu.be/wIaEaJGF0n8 - **Project:** https://cerebralvalley.ai/e/3rd-annual-natsec-hackathon/hackathon/gallery?project=43 ### 25. SmokenMirrors (Part 2) Acoustic Drone Detection model (i.e. detects drones via microphone and audio) - **Team:** [Drew Hooper](https://cerebralvalley.ai/u/dhooper) - **GitHub:** https://github.com/alexpiatek/Smoke-n-Mirrors-OS/tree/main/public - **Demo video:** https://nshackathon.palantirfoundry.com/shares/links/vlqxujk5nxypy - **Project:** https://cerebralvalley.ai/e/3rd-annual-natsec-hackathon/hackathon/gallery?project=94 ### 26. Hydra system Hydra is an autonomously command control system that is built to execute critical level missions autonomously, - **Team:** [Yeabsira Mulugeta](https://cerebralvalley.ai/u/YAB) - **GitHub:** https://github.com/xenn0010/hydrasys - **Demo video:** https://luma.com - **Project:** https://cerebralvalley.ai/e/3rd-annual-natsec-hackathon/hackathon/gallery?project=95 ### 27. K9 Sentinel – Nedron Labs K9 Sentinel is a rugged, portable, fully off-grid AI safety monitoring system for Military Working Dogs (MWDs). It provides real-time situational awareness inside vehicle kennels or crates using edge computer vision, environmental sensors, and intelligent alerts — no cloud, no cellular required. The Problem Heatstroke and undetected medical distress remain leading preventable causes of MWD fatalities. A February 2026 DoD Inspector General report documented 4 MWD deaths and 22 heat injuries over recent years due to inadequate kennel monitoring. Current solutions rely almost exclusively on basic temperature alarms and manual handler checks — insufficient in austere, disconnected environments (Humvees, MRAPs, helicopters, forward operating bases). Army doctrine (AR 190-12) requires portable airline-approved crates that move across any vehicle type, creating a unique gap no existing system fully addresses The Solution K9 Sentinel clips directly to any standard military crate and fuses: Computer vision (YOLO11/YOLOv8n-pose on NVIDIA Jetson Orin Nano Super) for posture/collapse detection, activity monitoring, and distress indicators. Environmental sensors (SHT31 temperature + humidity) for heat/humidity thresholds. Edge inference & alerting via local WiFi hotspot (or Meshtastic LoRa for true mesh off-grid). Everything runs onboard in real time. Handlers view a live dashboard on their phone/tablet showing camera feed with keypoint overlays, sensor readings, and instant alerts (Normal / Warning / Critical). The system is designed for contested, resource-constrained environments — exactly matching Army needs for resilient edge operations. Why It Fits the Hackathon Primary fit: Capability / Problem Statement 2 – Edge Deployments (lightweight edge computing, portable battery-powered kit, local inference, no cloud dependency). Secondary: Capability / Problem Statement 1 – Sensor Analysis & Integration (multi-modal fusion of vision + environmental data into actionable alerts). (Or Problem Statement 5: General National Security if you prefer — it still aligns strongly with the overall goals.) Built in <48 hours as a working prototype for the hackathon, it demonstrates technical feasibility and immediate military relevance. As an active SAR K9 handler myself, I designed this from real operational pain points. Impact Reduces preventable MWD casualties, improves handler safety and mission effectiveness, and provides a dual-use path (military + LE/DHS). Post-hackathon, we are ready for OTA/prototype funding and field pilots with units like the 341st Training Squadron. - **Team:** [Bryan Norden](https://cerebralvalley.ai/u/bryannorden) - **GitHub:** https://github.com/Nedron-Labs/K9-Sentinel - **Demo video:** https://youtu.be/GySrI3h7qvY - **Project:** https://cerebralvalley.ai/e/3rd-annual-natsec-hackathon/hackathon/gallery?project=6 ### 28. Cyber2Labs RF Sentinel is an AI-powered RF signal classification engine that treats every radio sensor as a data source to be analyzed and classified for operational prioritization. It streams spectrum data from Software-Defined Radios through a real-time pipeline that extracts spectral fingerprints, detects anomalies against learned baselines, and classifies emitters by type and threat level—enabling operators to distinguish routine traffic from jamming, spoofing, or unauthorized transmitters at a glance. Validated on Sub-GHz signals but frequency-agnostic across the entire spectrum, the system runs on commodity hardware from a $30 RTL-SDR to a $100 HackRF or USRP, making real-time spectrum intelligence and threat classification accessible beyond proprietary military-grade systems. - **Team:** [Ayyappan Rajesh](https://cerebralvalley.ai/u/iyapun), [Kevin Cardwell](https://cerebralvalley.ai/u/computerguru), [Day Nguyen](https://cerebralvalley.ai/u/dayn) - **GitHub:** https://github.com/nonamecoder/rf-sentinel - **Demo video:** https://youtu.be/i3FRUBpWlI0 - **Project:** https://cerebralvalley.ai/e/3rd-annual-natsec-hackathon/hackathon/gallery?project=34 ### 29. Icarus Prometheus (Icarus) is an offensive AI swarm for authorized security testing. It runs a multiagent system on top of LLMs and standard recon tools like subfinder, nuclei, and nmap, where each agent owns a role (recon, classifier, exploit, report) and writes back to a shared blackboard so new findings automatically trigger follow up work. The control panel is a Win95/XP themed local dashboard that shows every agent terminal, its prompts, responses, and tool calls in real time. The goal is to compress hours of manual probing into one scoped campaign you can watch end to end, with dry run on by default, so nothing fires without you saying so. - **Placement:** 5th Place - **Team:** [Nicolas Dos Santos](https://cerebralvalley.ai/u/nicolasDosSantos), [Francis Mendoza](https://cerebralvalley.ai/u/fmendoz7), [Nicholas Irving](https://cerebralvalley.ai/u/NickOne) - **GitHub:** https://github.com/NlCK01/prometheus-exe (updated) - **Demo video:** https://youtu.be/-Odnn4vHdpU (updated) - **Project:** https://cerebralvalley.ai/e/3rd-annual-natsec-hackathon/hackathon/gallery?project=35 ### 30. Watchertower Watchtower reduces the cybersecurity burden of delivering software to the U.S. government by automating the detection and remediation of critical vulnerabilities. It scans repositories with tools like Trivy, Grype, Gitleaks, and SonarQube to identify CVEs, leaked secrets, and code quality issues, then uses AI to analyze the affected code, generate fixes, run tests, and open pull requests for approved changes. By turning hours of manual vulnerability response, context switching, and validation into a single automated workflow, Watchtower helps military software teams stay secure while continuing to ship mission-critical features. - **Team:** [Ryan Brown](https://cerebralvalley.ai/u/rbrown), [Jeremy Losey](https://cerebralvalley.ai/u/jeremylosey), [Charles Peabody](https://cerebralvalley.ai/u/cpeabody) - **GitHub:** https://github.com/cj-cesi/watchtower - **Demo video:** https://youtu.be/ubj8zA7nfIk - **Project:** https://cerebralvalley.ai/e/3rd-annual-natsec-hackathon/hackathon/gallery?project=41 ### 31. Smoke n Mirrors OS Operator dashboard for fusing sensors detecting drones with acoustic, radar IR machine learning model - **Team:** [Alex Piatek](https://cerebralvalley.ai/u/ResQAP), [Ben Claflin](https://cerebralvalley.ai/u/benclaflin), [Drew Hooper](https://cerebralvalley.ai/u/dhooper), [Tony Adastra](https://cerebralvalley.ai/u/tonyadastra) - **GitHub:** https://github.com/alexpiatek/Smoke-n-Mirrors-OS.git - **Demo video:** https://youtu.be/6OWS3LLWpE0 - **Project:** https://cerebralvalley.ai/e/3rd-annual-natsec-hackathon/hackathon/gallery?project=97 ### 32. Team Squirtle Ghost-track is a drone detection, tracking, and response-deployment system that works from a live camera feed. It identifies aerial threats, tracks their movement in real time, and helps operators decide where to deploy countermeasures or security resources. Based on these detections, we also employ a world model inference engine based on the V-JEPA 2 video-foundation-model system for lost aerial-target re-acquisition [e.g. interceptor drone loses target drone performing evasive maneuvers around obstacles]. - **Team:** [vishruth bharath](https://cerebralvalley.ai/u/vishruthb), [Eddie Chen](https://cerebralvalley.ai/u/ayocheddie) - **GitHub:** https://github.com/vishruthb/ghost-track - **Demo video:** https://www.youtube.com/watch?v=1sQ9bzhPltc - **Project:** https://cerebralvalley.ai/e/3rd-annual-natsec-hackathon/hackathon/gallery?project=100 ### 33. Silver Dome The Silver Dome is a portable early-warning iron dome system that costs <$50 and can be deployed instantly to protect American military bases overseas, without the need for expensive missiles like Patriot. - **Team:** [Mukund Kunapareddy](https://cerebralvalley.ai/u/mukund), [Nicholas Perematko](https://cerebralvalley.ai/u/kinn), [Brandon Z](https://cerebralvalley.ai/u/Bran) - **GitHub:** https://github.com/Mukund2/Silver-Dome - **Demo video:** https://youtu.be/kNu1Q02rfUg - **Project:** https://cerebralvalley.ai/e/3rd-annual-natsec-hackathon/hackathon/gallery?project=7 ### 34. GhostNet Command & Control for Kill Web Management using 3D Gaussian splatting that allows you to send in a 360 drone to map out an entire combat zone, create hundreds or thousands of simulations for autonomous vehicle, train Reinforcement Learning Policies and finetune lora adapters that allow you to perfectly plan an engagement strategy with your autonomous weapons. - **Team:** [Brandon In](https://cerebralvalley.ai/u/brandonin) - **GitHub:** https://github.com/brandonin/national-security-hackathon - **Demo video:** https://youtu.be/O6_7b8qgKQE - **Project:** https://cerebralvalley.ai/e/3rd-annual-natsec-hackathon/hackathon/gallery?project=37 ### 35. Node We built a Command and control dashboard for visualizing constellations, with an AI chatbot layered in to autonomously execute on satellite collision avoidance. - **Team:** [Sidharth Anantha](https://cerebralvalley.ai/u/sidharthanantha), [Lundeen Cahilly](https://cerebralvalley.ai/u/lundeen), [Rahul Ayanampudi](https://cerebralvalley.ai/u/Rahulayanampudi) - **GitHub:** https://github.com/lundeen06/node - **Project:** https://cerebralvalley.ai/e/3rd-annual-natsec-hackathon/hackathon/gallery?project=38 ### 36. CommonGround We are fusing multimodal sensors to surface correlated threats. We are pulling in data from unmonitored sensors in the field, identifying signals above a threshold, generating SPOT reports from those signals, building a knowledge graph from those and other incoming SPOT reports, and continually identifying patterns, correlating threats, and creating pro-active alerts for the user. With an AI agent that can learn from the user's input on best practices and the ability to dynamically generate SITREPs. - **Team:** [Sean Knight](https://cerebralvalley.ai/u/thatsean), [Sreeram Venkatarao](https://cerebralvalley.ai/u/SreeramV181) - **GitHub:** https://github.com/SeanLikesData/CommonGround - **Demo video:** https://rubixcube.dubb.com/v/vcmx4r - **Project:** https://cerebralvalley.ai/e/3rd-annual-natsec-hackathon/hackathon/gallery?project=39 ### 37. Masa We built the fastest and most AI native mission control and command platform; voice activated with the ability to control a fleet of drones - **Team:** [Hasan Sukkar](https://cerebralvalley.ai/u/hasan_s), [Gufran Mirza](https://cerebralvalley.ai/u/gufranmirza) - **GitHub:** https://github.com/gufranmirza/natsec-hack-2026/tree/main - **Demo video:** https://www.loom.com/share/e1628942c55d49bebc707e73e3165bc6 - **Project:** https://cerebralvalley.ai/e/3rd-annual-natsec-hackathon/hackathon/gallery?project=46 ### 38. Package Backdoor Detection A black-box auditor for AI coding assistants that detects suspect post-training preferences for specific software packages — the supply-chain attack vector where an open-source LLM is fine-tuned to recommend a vulnerable npm package its developers know about and the public doesn't. We measure each model's post-training shift on a 600-package corpus and compare it against (a) a known-good calibration baseline (OLMo 2 1B base→Instruct) and (b) cross-suspect agreement across 4 different open-weights labs (DeepSeek, Qwen, InternLM, GLM). We find current signals are probably artifacts of the training data while identifying and demonstrating this novel attack vector; next steps involve finetuning specific models to better calibrate our outlier detection. - **Team:** [Rome Thorstenson](https://cerebralvalley.ai/u/rome) - **GitHub:** https://github.com/Rome-1/package-backdoor-detection - **Demo video:** https://www.loom.com/share/e227e62b182044be9ceec9713a1249f7 - **Project:** https://cerebralvalley.ai/e/3rd-annual-natsec-hackathon/hackathon/gallery?project=48 ### 39. Lincoln Lincoln — a governance runtime that lets autonomous weapon systems engage at machine speed while satisfying DoDD 3000.09's requirement for "appropriate levels of human judgment over the use of force." PROBLEM. Drone swarms have crossed the threshold where per-shot human authorization is operationally infeasible (Iran April 2024: $80M attack vs $1B defense; Red Sea: $1B in interceptors against $35K drones). The directive still demands transparent, auditable human judgment, but today's after-action investigations rebuild engagement records by hand over months — Tower 22 took 14 months; the USS Gettysburg blue-on-blue took nearly a year. SOLUTION. Three separately-accountable architectural layers: 1) Interpretation (pre-mission) — the commander's Rules of Engagement are parsed on-device, ambiguities surfaced for explicit resolution, and the compiled mission policy cryptographically signed. 2) Operation (engagement time) — a deterministic policy engine, with no LLM in the decision path, executes routine engagements inside the signed envelope at platform latency; exceptions route to operators. 3) Audit (continuous) — every decision is Ed25519-signed and hash-chained, traceable to the commander's authority, verifiable in ~41ms on commodity hardware without the runtime vendor's cooperation. Human judgment, applied once at compile time, scales to thousands of engagement decisions per mission. - **Team:** [L P](https://cerebralvalley.ai/u/lp0), [Toby Farmer](https://cerebralvalley.ai/u/Tobyfarm) - **GitHub:** https://github.com/tobyfarm/ArmyHackathon - **Demo video:** https://vimeo.com/1188853385/9e0e3e55f9?share=copy&fl=sv&fe=ci - **Project:** https://cerebralvalley.ai/e/3rd-annual-natsec-hackathon/hackathon/gallery?project=98 ### 40. Straw Hats Ghost Fleet Detector identifies ships that "go dark," intentionally disabling their AIS transponders to evade sanctions, smuggle cargo, or fish illegally. Every commercial vessel over 300 tons is required to broadcast its position, so we detect the absence of data rather than the presence. Our platform ingests AIS vessel tracking data, finds unexplained gaps in transmission, and scores each dark period using multiple factors: gap duration, distance traveled, proximity to known transshipment zones, implied speed, vessel type, and time of day. Scores are weather-adjusted using real storm data (NOAA) so legitimate outages don't generate false alerts. Flagged vessels are cross-checked against the OpenSanctions database, and suspicious locations can be verified with Sentinel Hub satellite imagery. Results are displayed on an interactive deck.gl map with risk charts, AI-generated intelligence summaries, and CSV export. Built with Next.js, Supabase, and OpenAI. - **Team:** [Aurelia Sindhunirmala](https://cerebralvalley.ai/u/americano), [Aldo Maximus](https://cerebralvalley.ai/u/Kamtomeh), [J S](https://cerebralvalley.ai/u/jarrensj) - **GitHub:** https://github.com/AureliaSindhu/ghostfleetdetector - **Demo video:** https://www.loom.com/share/6d8f12b8901c4041aa3d9e2fe1ca523d - **Project:** https://cerebralvalley.ai/e/3rd-annual-natsec-hackathon/hackathon/gallery?project=8 ### 41. Vector This is the Vector ISR Mission Planner. the goal of thi is to demonstrating a way of programming missions for drones that have decision points without having to have radio communications with the drone using the PEG-15 IR illuminator - **Team:** [Jarrod Sinclair](https://cerebralvalley.ai/u/Jarrod-Mythril), [Connor Sinclair](https://cerebralvalley.ai/u/ConnorS) - **GitHub:** https://github.com/jvsinclair/x-tech-hack-drone-mission-planning - **Demo video:** https://www.loom.com/share/0e9d8ebc374948a88e0f249098f7907d - **Project:** https://cerebralvalley.ai/e/3rd-annual-natsec-hackathon/hackathon/gallery?project=40 ### 42. HELLBAT - Acoustic Drone Detection HELLBAT — Acoustic Drone Detection at the Edge Modern counter-UAS depends on RF and radar — both fail against the threats that matter most. Acoustic is the missing modality. Two MEMS mics + ESP32-S3 sensor node feed a Jetson running bio-inspired modulation features and a probabilistic classifier; GCC-PHAT on stereo audio gives bearing. <200ms latency, no GPS, no RF emissions. Sensor node ~$20. Path to <$50/node at scale via custom SoC. A cheap node is a node you can lose, and a mesh of them triangulates not just drones, but small-arms fire and mortar teams. - **Team:** [Jared Childers](https://cerebralvalley.ai/u/absorbdmg) - **GitHub:** https://github.com/absorbdmg/edge-acoustic-detection - **Demo video:** https://youtu.be/sxLyOx9psMk?si=oyExMVQtM-yn1MJX - **Project:** https://cerebralvalley.ai/e/3rd-annual-natsec-hackathon/hackathon/gallery?project=9 ### 43. Vendrank One-click subcontractor ranking and flow-down document automation for defense primes, with a policy engine that guarantees FAR compliance and generates the sub-tier performance data CPARS misses.​​​​​​​​​​​​​​​​ - **Team:** [Andrew B McMorrow](https://cerebralvalley.ai/u/AndrewMcMorrow), [Lin Pan](https://cerebralvalley.ai/u/linpan), [Ari Ramsan](https://cerebralvalley.ai/u/aramsan) - **GitHub:** https://github.com/linpan0/CerebralValley - **Demo video:** https://youtu.be/Jeup1XLXEKU - **Project:** https://cerebralvalley.ai/e/3rd-annual-natsec-hackathon/hackathon/gallery?project=51 ### 44. Argus ARGUS ingests real-time Telegram signals from verified battlefield channels, enriches each event using an LLM to extract structured intelligence — actors, locations, equipment, casualties, faction attribution — and cross-verifies against GDELT media signals to produce a confidence-weighted event stream. The system exposes a unified command dashboard where analysts can scrub through historical battlespace evolution, watching events appear and transition from unverified to confirmed in real time as corroboration builds. A natural language query layer allows commanders to ask plain English questions — "what Russian equipment was reported near Kharkiv in the last six hours?" — and receive both the raw data and a structured SITREP in seconds. Every event is weighted by source trust, snapped to frontline position, and stored with full audit trail. The result is a living operational picture that compresses hours of manual intelligence synthesis into a continuously updated, queryable battlespace model. - **Team:** [Jeffrey Wang](https://cerebralvalley.ai/u/jeffreywangcf), [Myles Pember](https://cerebralvalley.ai/u/Mylespember) - **GitHub:** https://github.com/jeffreywangcf/argus - **Demo video:** https://youtu.be/huIZMSoXKfg - **Project:** https://cerebralvalley.ai/e/3rd-annual-natsec-hackathon/hackathon/gallery?project=54 ### 45. Sauron's Eye The goal is to develop an indoor reconnaissance drone that can navigate through buildings, scan its surroundings, and transmit a live video feed to a soldier’s AR goggles. Unlike a standard FPV drone, this system would detect and lock onto a target object, estimate its position within the building, and project that location into the soldier’s field of view. As the soldier moves, the system would continuously update the object’s relative position, allowing them to understand where it is, even when it is behind walls or outside their direct line of sight. basically active and small drones in soldier's backpacks and will deploy when the soldier need to clear out a trench, building and etc. Basically a wall hack where soldiers can shoot through walls. - **Team:** [Ezenbaatar Batjargal](https://cerebralvalley.ai/u/Ezen976), [Kunal Shrivastav](https://cerebralvalley.ai/u/poke-key), [Ruthwik Dasyam](https://cerebralvalley.ai/u/ruthwikdasyam) - **GitHub:** https://github.com/ruthwikdasyam/Saurons_eye - **Demo video:** https://drive.google.com/drive/folders/1M6Q8YddpfhAMHE_eounsxhAj9INTB-ro?usp=sharing - **Project:** https://cerebralvalley.ai/e/3rd-annual-natsec-hackathon/hackathon/gallery?project=99 ### 46. Somewear Labs Integrated edge-compute intel sharing solution for C2 and operator kill-chain decision synthesis - **Team:** [Mason Elms](https://cerebralvalley.ai/u/mason-somewear), [Matthew Roberts](https://cerebralvalley.ai/u/MatthewLRoberts), [Alan Besquin](https://cerebralvalley.ai/u/aBesquin) - **GitHub:** https://github.com/somewear-labs/xtech-deploy; https://github.com/somewear-labs/xtech-hackathon-android - **Demo video:** https://drive.google.com/file/d/1CPongp4RmhCDWxQFWAmS417deKpzEMlV/view?usp=sharing - **Project:** https://cerebralvalley.ai/e/3rd-annual-natsec-hackathon/hackathon/gallery?project=103 ### 47. Overpower Overpower is a predictive model of the fuel supply chain modeling over 187 different firm level agents, their responses to supply shocks and military deployment, and the impacts of various levers the DoW can use to procure and secure fuel and use the strategic petroleum reserve and defense production act to respond to supply shocks, engage in economic warfare, and alleviate harm to the civilian economy. - **Team:** [Arvin Sefidi](https://cerebralvalley.ai/u/arvisef) - **GitHub:** https://github.com/Asefidi/Overpower - **Demo video:** https://youtu.be/8K2bRj1-1RQ - **Project:** https://cerebralvalley.ai/e/3rd-annual-natsec-hackathon/hackathon/gallery?project=10 ### 48. graph Manifold is a first-of-its-kind geometric coordinate system for information warfare. While frontier models like ChatGPT-5.5 Pro hit a 'structural scaling wall' and perform like random noise at scale, manifold uses a continuous-time neural ode architecture to navigate the folds and friction of a contested environment. By modeling the hidden geometry of community boundaries, the manifold, we've created a resilient coordination layer that outperforms SOTA (ChatGPT 5.5 Pro) with a 21x coverage lead, 1.5x tracking error reduction in error, and 120x more total graph control. We aren't just simulating the cascade; we're maneuvering through it to achieve true decision dominance. Smarter, not bigger. - **Team:** [Aditya Mangalampalli](https://cerebralvalley.ai/u/Aditya2162) - **GitHub:** https://github.com/amangalampalli/manifold - **Demo video:** https://www.youtube.com/watch?v=-SvRJzEgcHA - **Project:** https://cerebralvalley.ai/e/3rd-annual-natsec-hackathon/hackathon/gallery?project=42 ### 49. Mighty Morphing Mighty Morphing — RealID for Voice Authority Why AI voice cloning has collapsed voice as a chain-of-command primitive. A senior officer’s voice can now be cloned from 30 seconds of a press briefing, podcast, or voicemail greeting using consumer-grade tools (ElevenLabs, Cartesia, OpenAI). Voice-authorized workflows — release authorities, strike-package approvals, financial wire approvals, AI agent voice interfaces — were built on the assumption that “I recognize their voice.” That assumption is dead. Evidence Arup (2024): $25M wire-transfer fraud via deepfake video call impersonating the CFO FBI: ~1,000 voice-clone scams reported per quarter Pentagon-adjacent risk: deepfake of a senior leader could authorize lethal action, release classified data, or redirect logistics before human verification catches up Existing voice biometric vendors (Pindrop, Nuance) cannot cryptographically bind audio to identity — and voice biometric alone is exactly what clones are designed to beat Use Cases Voice-authorized command verification — duty officer authorizations gated by signed audio + voiceprint + deepfake check before agents act AI agent voice gating — autonomous agents operating on behalf of senior officials require cryptographically-signed audio inputs, not just transcribed words Clearance-grade voice authentication — replaces single-factor voice biometric with three independent failure modes Forensic voice verification — any captured audio can be verified against the public registry without trusting the verifier Our Solution Mighty Morphing is a public trust gate for voice. An enrollment center (analogous to RealID) captures the voice sample, derives the ECAPA-TDNN voiceprint, and signs a public attestation: public key + voiceprint fingerprint + issuer signature + status. Every voice-bound action is then gated by three independent defenses: Cryptographic signature — Ed25519 binds an identity to specific audio (SHA-512). Captured signatures cannot be replayed on substituted audio. Voiceprint match — ECAPA-TDNN cosine ≥ threshold against the registry’s enrolled embedding catches impersonation. Deepfake detection — Wav2Vec2 + WavLM ensemble (production v13 head: 99.14% TPR on ElevenLabs, 98% on Cartesia, sub-1% real-speaker false positive) catches synthesis even when voiceprint matches. Defeating all three simultaneously requires the attacker to hold the device with the private key and clone the voice convincingly and keep deepfake risk below threshold. No single defense is sufficient; the design constraint is all three failing at once. What’s Live in the Demo Public registry with lookup by ID / fingerprint / public key Identity bundle separated from public envelope — private key delivered exactly once, browser-cached, never returned by registry reads Live Cartesia voice cloning of the operator’s own recording, downloaded and submitted to the verifier, and rejected by the layered defense in real time WebAuthn passkey ceremonies (backend ready) and ML-DSA-65 post-quantum signatures (roadmap) so the private key can be device-sealed and signatures remain valid through CRQC Stack: Go gateway · SpeechBrain ECAPA-TDNN voiceprint sidecar · Wav2Vec2 + WavLM deepfake ensemble · React/TypeScript UI · Cartesia voice-clone API · Ed25519 → ML-DSA-65 / WebAuthn roadmap - **Team:** [Johnny Hung](https://cerebralvalley.ai/u/masterfung) - **GitHub:** https://github.com/masterfung/natsec-hackathon - **Demo video:** https://youtu.be/ghAgrFYzwB4 - **Project:** https://cerebralvalley.ai/e/3rd-annual-natsec-hackathon/hackathon/gallery?project=52 ### 50. SwarmD2C2 defensive drone swarm command and control for the tactical edge. - **Team:** [Scott McCutche](https://cerebralvalley.ai/u/Soverance), [Natalie Rakoski](https://cerebralvalley.ai/u/KINETAS), [Sahilkrishna Vazhathodiyil](https://cerebralvalley.ai/u/sahilkv), [Laszlo Szilagyi](https://cerebralvalley.ai/u/Laszlo) - **GitHub:** https://github.com/KINETAS-CO/swarmd2c2 - **Demo video:** https://youtu.be/3WjnDziezgU - **Project:** https://cerebralvalley.ai/e/3rd-annual-natsec-hackathon/hackathon/gallery?project=53 ### 51. Project Tessera Project Tessera is an adversarial decision-making substrate that lets frontier AI models plan, execute, and learn from battle plans without retraining. A hex-grid simulator with fog-of-war and multi-modal sensor fusion handles the battlespace; Claude Sonnet 4.6 plans every turn via tool-use, surfacing per-unit reasoning the operator can audit. After each game, Claude Haiku 4.5 reads the trace and distills two to four lessons into an append-only memory file that feeds straight back into the next game's context. The same model gets better because its experience compounds. We support human vs AI, AI vs AI, and hot-seat play, with drag-and-drop placement and a probabilistic intel picture you can click into. - **Team:** [Alexandre Carlhammar](https://cerebralvalley.ai/u/acarlham), [Brian Check](https://cerebralvalley.ai/u/briancheck9) - **GitHub:** https://github.com/b9check/NSA-Hackathon - **Demo video:** https://docs.google.com/presentation/d/13Hh6MXvV_Mp2NFXyoHzhZDoVp1tqYvjSF3v86Ni0gfs/edit?usp=share_link - **Project:** https://cerebralvalley.ai/e/3rd-annual-natsec-hackathon/hackathon/gallery?project=55 ### 52. ARGUS-ORACLE ORACLE — Objective Reasoning And Cognitive Liability Engine — is an intelligence failure prevention platform that attaches one detector to each of the five recurring cognitive failure modes behind every modern intelligence catastrophe (9/11, Iraq WMD, Afghanistan 2021, October 7, Teixeira/Rahman). Six integrated modules — PHANTOM (5-dimension AIS vessel threat scoring), CASSANDRA (immutable warning ledger with 48-hour timer and Claude-graded convergence detection), DEAD STAR (Bayesian assumption decay that blocks operational decisions when red), TELEPHONE (brief-vs-source distortion auditor), ORACLE Core (Brier-ranked analyst leaderboard with cognitive bias fingerprint and internal prediction market), and VISION (live image/video/satellite analysis via TF.js + Qwen3-VL) — ride on a Palantir Foundry AIP-native ontology with 7 Object Types, 10 Link Types, and 8 typed Action Types, every mutation audit-logged. Reasoning runs on Claude Opus 4.7 with adaptive thinking; data flows from 36 free open-source feeds (aisstream.io, GDELT, ACLED, OFAC SDN, NOAA GOES-16, NASA Worldview, Polymarket, OpenSky, USGS, Cloudflare Radar, etc.) at $0/month operating cost. The hero proof: today, with a real Anthropic key, ORACLE's TELEPHONE module scored the actual 2003 DIA→Powell Iraq WMD pair at 92/100 distortion and blocked delivery — citing the State and DOE dissents and the Curveball source-validation chain that the IC missed in real time. - **Team:** [Kush Ise](https://cerebralvalley.ai/u/KUSH2704) - **GitHub:** https://github.com/Kush614/oracle-natsec-hackathon - **Demo video:** https://www.loom.com/share/fdd4587794bd4a56afce6051f57c739b - **Project:** https://cerebralvalley.ai/e/3rd-annual-natsec-hackathon/hackathon/gallery?project=101 ### 53. Ryan_4_Ryan High resilience system to maintain target custody in contested airspace. It uses sensor fusion and multi-arm bandit to select the best sensing and tracking strategy. Works really well for targets having abrupt maneuvers/trajectories. - **Team:** [Ryan Chekkouri](https://cerebralvalley.ai/u/Ryan_MLP) - **GitHub:** https://github.com/Ryan-NLP/natsec_hackathon - **Demo video:** https://youtu.be/-bdvEgqXBfM - **Project:** https://cerebralvalley.ai/e/3rd-annual-natsec-hackathon/hackathon/gallery?project=11 ### 54. Team name: VTS (Vector to Semantics) for Drone EDGE AI operation In bandwidth-constrained environments, streaming raw drone video overwhelms both networks and operators. With the VTS protocol, we send less—but infer more—using edge AI to deliver only meaningful insights. - **Team:** [Shawn Park](https://cerebralvalley.ai/u/foozim), [David Choi](https://cerebralvalley.ai/u/ai4ce), [Taehee Yoon](https://cerebralvalley.ai/u/jadenyoon), [Edward Kim](https://cerebralvalley.ai/u/maeng31) - **GitHub:** https://nshackathon.palantirfoundry.com/workspace/compass/view/ri.compass.main.folder.df2b1955-b140-48e2-99a0-23085e855c8c - **Demo video:** https://youtu.be/itFSr2VlJ4E - **Project:** https://cerebralvalley.ai/e/3rd-annual-natsec-hackathon/hackathon/gallery?project=44 ### 55. Ghostline-Blackout Ghostline is an edge AI field copilot for military operators working in denied, contested, or comms-degraded environments where cloud tools, GPS, and command connectivity can fail. It runs on phone-class hardware and combines camera CV, local intelligence, an encrypted event log, HMAC-signed sync, and a retrieval-bound multi-voice AI council to keep operators survivable, supervised, and traceable when the signal disappears. The system delivers a 9-page operation dashboard with five operation native intelligence views: Hub, Constellation, Flow, Layers, and Risk Stack. It also includes a Cesium Ion Photo3D operational picture and an operation-bound voice copilot powered by OpenAI Realtime, ElevenLabs multi-voice TTS, and Web Speech STT. Every reply is evidence cited, ROE bound, and reviewed by a six agent council consisting of Navigator, Safety, Comms, Cyber, Intel, and Guardrail, with each agent speaking in its own ElevenLabs voice. A hard safety boundary in every prompt rejects targeting, attack planning, weapons guidance, and hostile-intent inference. Voice is wired into six pages with page-specific tools that can pan the map to an operator, narrate briefing sections, scroll to events, and describe terrain. We shipped a runnable prototype with 13 operations across four classification tiers, 8 operator profiles, a complete Foundry and Ontology data model, offline-first edge execution, encrypted local logging, and integrity-checked delta sync that lets the command reconstruct the entire offline window when connectivity returns. - **Team:** [Ayush Ojha](https://cerebralvalley.ai/u/ayushojha) - **GitHub:** https://github.com/ayushozha/ghostline - **Demo video:** https://youtu.be/Qwzz3qtqo1w - **Project:** https://cerebralvalley.ai/e/3rd-annual-natsec-hackathon/hackathon/gallery?project=102 ### 56. EnQuanta EnQuanta's Dynamic Hybrid Crypto-Agility Framework addresses a growing operational challenge: cryptographic algorithms do not last forever, and systems that hardcode encryption require costly re-engineering whenever vulnerabilities emerge. Current post-quantum cryptography (PQC) solutions do not adequately address the harvest-now, decrypt-later (HNDL) threat, particularly for long-lived data with national security implications. This demo presents a concrete approach: organizing algorithms into versioned cipher pools that operators can update, review, and publish without modifying application code. Both symmetric and asymmetric pools are supported, with the asymmetric pool combining classical and NIST-standardized post-quantum algorithms side by side. A live messaging session demonstrates the framework in action, deriving session keys from the active pools and displaying the full cipher stack for each message to ensure auditability. - **Team:** [William Austin](https://cerebralvalley.ai/u/william-enquanta), [Tucker Saxon](https://cerebralvalley.ai/u/tsaxenq), [Eric Mencke](https://cerebralvalley.ai/u/ericmencke) - **GitHub:** https://github.com/enquanta/enquanta-ns-hackathon - **Demo video:** https://app.guidde.com/share/playbooks/5SGN31v5e5iv7MDgywYGjU?origin=qjfgxjkTUgMYH8TDN5OO1O76YdI3&mode=videoOnly - **Project:** https://cerebralvalley.ai/e/3rd-annual-natsec-hackathon/hackathon/gallery?project=12 ### 57. robot SAx is a dual use drone dispatch and control program. Disaster response and national security teams often work in places where normal infrastructure is degraded or unavailable. They may have limited bandwidth, limited power, too many live feeds, and too few people watching them. SAx turns raw drone observations into structured mission intelligence out of the backpack without the need for power or internet. - **Team:** [Martin W](https://cerebralvalley.ai/u/robotmartin) - **GitHub:** https://github.com/wojtczyk/SAx - **Demo video:** https://youtube.com/shorts/IsId_p5hYzg - **Project:** https://cerebralvalley.ai/e/3rd-annual-natsec-hackathon/hackathon/gallery?project=13 ### 58. FOOTPRINT https://footprint.army Pre-deployment adtech self-audit tool. Measures and mitigates advertising technology surveillance exposure at the personal, unit, and site level. Measure and mitigate your advertising technology surveillance exposure before deployment. Framework for Observability, Optimization & Tracking of Privacy, Risk, Identity & Network Telemetry - **Team:** [Lucas Pick](https://cerebralvalley.ai/u/drduker), [Eric Wardner](https://cerebralvalley.ai/u/noteric), [Joshua Taylor](https://cerebralvalley.ai/u/jtaylor) - **GitHub:** https://github.com/tiberius-grail/FOOTPRINT - **Demo video:** https://youtu.be/A03ixPHc1vo - **Project:** https://cerebralvalley.ai/e/3rd-annual-natsec-hackathon/hackathon/gallery?project=14 ### 59. H.A.D.E.S: Hierarchical Autonomous Drone Edge System HADES is a security first drone swarm simulation that trains edge aware UAV teams to protect convoys, detect threats, reroute safely, and support humanitarian response when networks degrade. - **Team:** [aTG R](https://cerebralvalley.ai/u/atg) - **GitHub:** https://github.com/r-agni/hades - **Demo video:** https://drive.google.com/drive/folders/1tqkZo6x4NlKAUxYh44Ee3nqVZgzYR7yh?usp=sharing - **Project:** https://cerebralvalley.ai/e/3rd-annual-natsec-hackathon/hackathon/gallery?project=17 ### 60. SplatSwarm SplatSwarm is a terrain-aware autonomous drone swarm mission rehearsal engine that lets commanders stress-test missions before any drone leaves the ground. Operators inject real battlefield failures such as GPS jamming, comms blackout, mid-mission drone losses, and watch the swarm adapt in real time over photorealistic satellite terrain. Every drone movement, sensor event, and human kill chain approval is logged immutably to Palantir Foundry, generating the ground truth dataset that trains reinforcement learning models for smarter, more resilient future missions. Same hardware, better software, zero surprises over the target because the drones have already seen every possible scenario. - **Team:** [Umar Ghani](https://cerebralvalley.ai/u/UmarGhani), [Steve Kuo](https://cerebralvalley.ai/u/stevedusty), [Sanjay Marathe](https://cerebralvalley.ai/u/itsmesmarathe), [Mohul Shukla](https://cerebralvalley.ai/u/mohulshukla) - **GitHub:** https://github.com/Steve-Dusty/drone-swarms - **Demo video:** https://www.youtube.com/watch?v=TtrOXJIfPUs - **Project:** https://cerebralvalley.ai/e/3rd-annual-natsec-hackathon/hackathon/gallery?project=23 ### 61. multi signal Fetching multi signals from camera feeds and network logs to capture anomalous behaviors - **Team:** [Roey Ben Chaim](https://cerebralvalley.ai/u/roeybc), [Boris Ploix](https://cerebralvalley.ai/u/swrd06bp) - **GitHub:** https://github.com/JaiJai060708/HackCam - **Demo video:** https://drive.google.com/file/d/1r6I2hSa5jB3pk0taZPwHMBUzSJkmJ6yI/view?usp=sharing - **Project:** https://cerebralvalley.ai/e/3rd-annual-natsec-hackathon/hackathon/gallery?project=57 ### 62. Project ARGUS Project ARGUS draws from previous high-altitude solar gliders to position forward sensors in any domain to detect hypersonic glide missiles. The glider features edge computing, which contains a custom-designed AI model architecture to sense differential air pressure, IR and EO to locate and track hypersonic threats. - **Team:** [Andy Liu](https://cerebralvalley.ai/u/andwinliu), [Liam Collins](https://cerebralvalley.ai/u/LiamCollins), [Mihir Sharma](https://cerebralvalley.ai/u/mihirsharma), [Krishna Kesavalu](https://cerebralvalley.ai/u/studentkrisk) - **GitHub:** https://github.com/AndyLiu0/national-security-hackathon - **Demo video:** https://youtu.be/DwH9sXS7Pyc - **Project:** https://cerebralvalley.ai/e/3rd-annual-natsec-hackathon/hackathon/gallery?project=60 ### 63. Aias Shield Aias Shield is a mission-alignment layer for autonomous AI systems that helps prevent models from acting on compromised, low-trust, or misleading external context. As AI agents become more autonomous, they increasingly rely on live inputs like maps, reports, sensor feeds, metadata, retrieved documents, and tool outputs. A single poisoned context item can quietly shift the model away from the operator’s original intent and cause it to make a confident but unsafe recommendation. Aias Shield solves this by checking whether an AI system is still aligned before it acts. It anchors the mission intent, inspects incoming context, compares trusted baseline reasoning against context-influenced reasoning, detects intent drift, blocks unsafe outputs, preserves a safe fallback, and produces an evidence trail for human review. - **Team:** [Alex Chao](https://cerebralvalley.ai/u/alexchaomander) - **GitHub:** https://github.com/alexchaomander/aias-shield - **Demo video:** https://youtu.be/FuglaSDfCPw - **Project:** https://cerebralvalley.ai/e/3rd-annual-natsec-hackathon/hackathon/gallery?project=15 ### 64. Callout by Ozni AI Callout is a dialogue-based command and control (C2) protocol that enables Machine-to-Machine-to-Operator collaboration with voice as the substrate. When dismounted and expeditionary tactical units are collaborating with autonomous warfighting systems the common operating picture (COP) gets left behind, Callout ensures that team commanders, team members, UxS, and sensors alike can all participate in the kill-chain even in disconnected and denied environments. Our demonstration illustrates the use of Callout when an operator needs to coordinate a ground sensor collection, a UAS flight plan, query previous data stored in the Palantir Foundry AIP, and issue commands by speaking into their radio. Each autonomous system enabled by Callout: 1. Receives the RF transmission and Demodulates the signal 2. Transcribes the voice message with OpenAI's Whisper 3. Performs intent-based reasoning using a locally deployed LLM 4. Executes commands or fulfills data requests internally 5. Responds with acknowledgements and asks for clarification if needed 6. Coordinates with other UxS assets to achieve stated objectives To build Callout, hardware we used includes an NVIDIA Orin AGX, two HackRF One Software defined radios, and a RemoteID transmitter. Software products included OpenAI's Whisper model, Palantir's Foundry AIP, Paige Text-To-Speech for on-device voice generation, and locally hosted LLM models for voice and text inference. The Callout protocol will fundamentally shift the way tactical operators think about collaborating with autonomous warfighting systems. By giving sensors and UxS a voice, they can participate, collaborate, and protect warfighters at the tactical edge. Whether for force protection or force multiplication, the Callout protocol will be the catalyst for widespread adoption of autonomous intelligent systems at the tactical edge. - **Team:** [Dominick Perini](https://cerebralvalley.ai/u/perindom), [Justin Kopacz](https://cerebralvalley.ai/u/jkopacz) - **GitHub:** https://github.com/jkopacz-ozni/NSH - **Demo video:** https://youtu.be/_TKNKVJACZM - **Project:** https://cerebralvalley.ai/e/3rd-annual-natsec-hackathon/hackathon/gallery?project=16 ### 65. Skunk Works v2 We built a simulator to gauge the effects of environmental factors on the integrity of autonomous vehicles (AV). Our platform empowers engineers to understand how different materials and components hold up in an environment, giving them the insight needed to build better products. As for the post production lifecycle of AVs, we provide operation and maintenance teams the tools to monitor how their equipment handles everything from dust and heat to salt and ice. - **Team:** [Sabal Poudel](https://cerebralvalley.ai/u/sabal), [Vatsal Srinet](https://cerebralvalley.ai/u/Virginia-tech), [Enzo Nunez](https://cerebralvalley.ai/u/enzonunez), [Shashvath Srivatsa](https://cerebralvalley.ai/u/shashvaths) - **GitHub:** https://github.com/Sabalpp/HackSEC_SF - **Project:** https://cerebralvalley.ai/e/3rd-annual-natsec-hackathon/hackathon/gallery?project=21 ### 66. TacNet TacNet Edge is a low-bandwidth mission command system that uses local small language models to convert voice reports into text, extract mission- critical signals, remove unnecessary noise, and compress the result into compact binary metadata for transmission across degraded mesh networks. Instead of sending raw audio or long transcripts, TacNet transmits only the operational meaning: who reported, what happened, where it happened, urgency, unit status, personnel readiness, equipment status, and requested action. This allows critical reports such as 9-line/CASEVAC, SALUTE, ACE/ LACE, resupply, and sensor observations to move through bandwidth-constrained links that would fail under normal audio transmission. On the receiving side, the MI/S2 dashboard decodes the binary metadata, reconstructs the report, updates personnel and asset readiness, and uses an embedded LLM layer to help analyze incoming reports from soldiers, vehicles, drones, and sensors. The goal is to reduce information overload and help commanders make faster, more informed decisions under contested communications conditions. - **Team:** [Ethan Joffe](https://cerebralvalley.ai/u/xtrek), [Vinal Chaudhary](https://cerebralvalley.ai/u/CyberCthulu), [yifu zuo](https://cerebralvalley.ai/u/yifu001), [James Vo](https://cerebralvalley.ai/u/jamjam112) - **GitHub:** https://github.com/ethanj/tacnet - **Demo video:** https://www.youtube.com/watch?v=_8_IktRLphs - **Project:** https://cerebralvalley.ai/e/3rd-annual-natsec-hackathon/hackathon/gallery?project=58 ### 67. Team Spire C2D2 is a combat decision dominance platform built on TAK to be configured on ATAK for soldiers undergoing Ranger School Training. It supposed to digitize the data gathering process from writing in books to using audio recordings that analyze a cadets performance on the go. It also helps with Squad building and Squad simulations against opposing forces to help leaders assemble the best teams possible - **Team:** [Daniel Yeboah](https://cerebralvalley.ai/u/dy_official), [Fresnel Fabian](https://cerebralvalley.ai/u/Fresnel), [Susana Owusu-Ansah](https://cerebralvalley.ai/u/Suez_ann), [Godliver Alangyam](https://cerebralvalley.ai/u/Liver) - **GitHub:** https://github.com/legal-ai-2026/Spire-Frontend - **Demo video:** https://www.youtube.com/watch?v=KiVgE0BwmaA - **Project:** https://cerebralvalley.ai/e/3rd-annual-natsec-hackathon/hackathon/gallery?project=59 ### 68. Ghostline Voice-driven cascade intelligence on Palantir Foundry. - **Team:** [Sanjay Sai](https://cerebralvalley.ai/u/sanjay_sai), [Neer Aeron](https://cerebralvalley.ai/u/Neer), [Abhiram Yadu Sribhashyam](https://cerebralvalley.ai/u/asribhas), [Peyton Li](https://cerebralvalley.ai/u/peli) - **GitHub:** https://github.com/abhirams303/ghostline - **Demo video:** https://www.loom.com/share/8af7963405e341f6bfecb66128f06172 - **Project:** https://cerebralvalley.ai/e/3rd-annual-natsec-hackathon/hackathon/gallery?project=64 ### 69. STAT-TIS Glass We built a wearable, edge-based system that turns real-world observation into structured data—automatically capturing personnel and supply counts without manual reporting. This enables real-time visibility and faster, more accurate decision-making in operational environments where traditional reporting is slow, error-prone, and unreliable. - **Team:** [Josh Sembrano](https://cerebralvalley.ai/u/Joshsembrano), [Will Viana](https://cerebralvalley.ai/u/willViana), [Johnny Hunter](https://cerebralvalley.ai/u/nommminal) - **GitHub:** https://github.com/vdawger/statis-glass - **Demo video:** https://drive.google.com/drive/folders/1jnJHK7F67TcNsberRQ5GERd7n3m6enSC - **Project:** https://cerebralvalley.ai/e/3rd-annual-natsec-hackathon/hackathon/gallery?project=81 ### 70. TruePoint The Tactical Edge Route Agent (TERA) : A voice driven, offline AI agent for tactical operators, running entirely on a Jetson Orin Nano. Operator speaks or types intent into an on-device LLM (Gemma), which then creates a ML-DSA-65 PQC signed response, and then has scaffolding for an on-device TTS guided rationale for tactical route guidance through a curated operator-cadence voice profile tailored from open voice models. A six-stage security pipeline (prompt guard, redaction, provenance, schema validation, policy gate, trust scoring) that rejects prompt injection before the model on prem sees it. ATAK plugin connects the operator device to the Jetson Nano over LAN, where the verify-gate refuses unsigned or untrusted key-responses before rendering. - **Team:** [Jonathan Goohs](https://cerebralvalley.ai/u/jon-goohs), [Kyle Hicks](https://cerebralvalley.ai/u/kyle_hicks), [Ben Schwierking](https://cerebralvalley.ai/u/Ben_Schwierking), [Satriyo utomo](https://cerebralvalley.ai/u/aleens) - **GitHub:** Https://github.com/jdev-02/tera - **Demo video:** https://youtu.be/HosWl0E4u0I?si=pWSW-ljMtUbOJ3bA - **Project:** https://cerebralvalley.ai/e/3rd-annual-natsec-hackathon/hackathon/gallery?project=82 ### 71. RI-AI (Harvard QLab) Today's command-and-control platforms are built for staff and battle captains, and they're only as good as the input they get. Frontline units don't plan digitally: they plan with paper maps and acetate, then communicate sparse summaries via radio up the chain. Higher headquarters have a fraction of the picture they could. This project gives small-unit leaders an intuitive mission-planning tool that makes digital planning faster than paper-based planning and pushes rich frontline data up to higher headquarters to increase their awareness of the battlefield. - **Team:** [Nils Olsen](https://cerebralvalley.ai/u/chillsberry), [Young Jae Koh](https://cerebralvalley.ai/u/YoungJaeContext) - **GitHub:** https://github.com/hi-aduek/RIAI - **Demo video:** https://youtu.be/BLed4WswNp8 - **Project:** https://cerebralvalley.ai/e/3rd-annual-natsec-hackathon/hackathon/gallery?project=84 ### 72. SkyCustody SkyCustody is an edge-first SIGINT pipeline that solves the mentor's exact ask: a 149 KB CNN + autoencoder on-device turns 232 GB/hr of raw RF into ~600 B intelligence rows. Each event flows through a custody state machine, TDOA geolocation (Chan 1994, ~30 m CEP), blue-force attribution, and a streaming-DBSCAN persistence detector that flags "I can't explain this one." Every output streams live into 7 typed datasets on a real Palantir Foundry tenant via the publishRecord API, with a DDIL JSONL replay buffer so the edge node never loses data when comms drop. The full Find→Fix→Track→Target kill chain closes in one tactical dashboard with one-click COA execution. - **Team:** [Rajashekar V](https://cerebralvalley.ai/u/raj) - **GitHub:** https://github.com/rajashekarcs2023/national-security-3 - **Demo video:** https://youtu.be/T3wNpKiv6BM - **Project:** https://cerebralvalley.ai/e/3rd-annual-natsec-hackathon/hackathon/gallery?project=86 ### 73. Oceans of Vectors GPS denied autonomous budget friendly quad copter drones (Nvidia Isaac robotics simulator environment) - **Team:** [Kyle Sorensen](https://cerebralvalley.ai/u/kyles) - **GitHub:** https://github.com/oceansofvectors/azimuth - **Demo video:** https://www.youtube.com/watch?v=meS70HpjuhM - **Project:** https://cerebralvalley.ai/e/3rd-annual-natsec-hackathon/hackathon/gallery?project=18 ### 74. AdaptSim Military and industrial training is still constrained by two brittle modes. Physical simulation is expensive, fixed in place, and hard to reset. Live-fire ranges, wet trainers, mockups, and rehearsals can be valuable, but they usually produce limited repetitions against a small number of preplanned situations. Digital simulation has the opposite problem. It is easier to repeat, but useful scenarios still require teams of designers, engineers, and subject matter experts to author specific environments, assets, decision trees, and failure modes in advance. The long tail of real-world situations is rarely represented. The result is a gap between the schoolhouse and the field. Training often certifies that a team can pass a known standard, but it does not always expose how individuals and teams adapt under ambiguous, changing, high-pressure conditions. AdaptSim is aimed at that gap: quickly turn a real environment into a reusable training space, then generate doctrine-informed adversarial scenarios that vary across repetitions while remaining physically grounded in the scanned space. - **Team:** [Nick Parkes](https://cerebralvalley.ai/u/nick_parkes), [Mackenzie Lloyd](https://cerebralvalley.ai/u/Mackenzie-Lloyd) - **GitHub:** https://github.com/nickparkes7/adaptsim-hackathon - **Demo video:** https://www.loom.com/share/277f5654db6f42ed8a00a99a47fa1dff - **Project:** https://cerebralvalley.ai/e/3rd-annual-natsec-hackathon/hackathon/gallery?project=61 ### 75. BattleSpace BattleSpace C2 is a single‑screen Common Operating Picture (COP) for a small operations cell. It fuses live AIS, simulated EO/IR/RF/radar, and a real or simulated drone feed onto a tactical map (MapLibre + deck.gl), and orchestrates a four‑agent crew (fusion‑agent, sensor‑tasker, intel‑analyst, drone‑operator) through the F2T2EA kill‑chain (Find → Fix → Track → Target → Engage → Assess) using the gastown multi‑agent coordinator. Every operator action becomes a real gt command (convoy create, broadcast, synthesis add, mq approve, wl post), with a Hold‑to‑Authorize human‑in‑the‑loop gate before any engagement. The result: an auditable, sub‑minute kill‑chain loop that accelerates Mission C2 while keeping a human on the trigger. - **Team:** [Tony Loehr](https://cerebralvalley.ai/u/forkbomb), [William Detscher](https://cerebralvalley.ai/u/willdetscher) - **GitHub:** https://github.com/jl-codes/battlespace-c2 - **Demo video:** https://youtu.be/t0hNhxoy2Hc - **Project:** https://cerebralvalley.ai/e/3rd-annual-natsec-hackathon/hackathon/gallery?project=69 ### 76. JTAC Loop We automated akey portion of the air-support to JTAC to Ground Force Commander kill-chain, in which the JTAC operator develops a 9-line description of the air support request. Normally, this takes 3-5 minutes due to the high amount of verbal and manual control required from the operator. Our system recommends an action for the JTAC to approve. This shortens the operation to less than a minute, and is fully integrated into current ATAK workflows. - **Team:** [Luca Violone](https://cerebralvalley.ai/u/LucaViolone), [Olivia Taylor](https://cerebralvalley.ai/u/otaylor), [Collen David Larson](https://cerebralvalley.ai/u/Collen), [will johnson](https://cerebralvalley.ai/u/willjohnson) - **GitHub:** https://github.com/otaylor2023/JTAC_LOOP - **Demo video:** https://drive.google.com/drive/folders/1SqSUcMaGl6oX8OAT9xHHCPmjU1VGrg7Z?usp=sharing - **Project:** https://cerebralvalley.ai/e/3rd-annual-natsec-hackathon/hackathon/gallery?project=72 ### 77. Expanso Distributed data pipelines at the edge. - **Team:** [David Aronchick](https://cerebralvalley.ai/u/aronchick) - **GitHub:** https://github.com/aronchick/expanso-edge-video-and-event-detection - **Demo video:** https://share.descript.com/view/5Wl9fRwMWMK - **Project:** https://cerebralvalley.ai/e/3rd-annual-natsec-hackathon/hackathon/gallery?project=75 ### 78. Crater-Be-Gone LOE 2: Drone Ops Conducting rapid Battle Damage Assessment during JADR Operations - **Team:** [Hans Fotta](https://cerebralvalley.ai/u/hansf16), [Ariana Canfield](https://cerebralvalley.ai/u/Canfield), [Daniel Osborne](https://cerebralvalley.ai/u/DOsborne), [Lars Lofquist](https://cerebralvalley.ai/u/LarsLofquistSPK) - **GitHub:** https://github.com/DanielTOsborne/DroneBDA/tree/main - **Demo video:** https://www.youtube.com/watch?v=VEP1VgNYO9E - **Project:** https://cerebralvalley.ai/e/3rd-annual-natsec-hackathon/hackathon/gallery?project=90 ### 79. Skylock Skylock is the trust layer for Counter-UAS. Every drone-detection product on the market answers one question: is there a drone? Skylock answers a different one: was this defensible? Each detection produces a signed, hash-chained GroundTruth Receipt — the same evidence shape a JAG officer would accept in a use-of-force review. Skylock fuses acoustic, optical, RF Remote ID, and IMU votes through SCI v1: a seven-component custody-inference algorithm that feeds a deterministic seven-state state machine. A Trust Gate verifies the model engine’s SHA256 against a signed passport before every inference. If the model is tampered with, Skylock does not fail silently. It still produces a signed receipt, but trust is automatically adjusted and the audit chain logs the gate trip. A Decoy Verifier uses parallax and temporal motion checks to catch printed-photo and replayed-video spoofs. The operator console runs in any browser — laptop, phone, wall display, or offline tablet. The NodePack runs on-device only: Jetson Orin Nano, with YOLO11n FP16 inference at roughly six milliseconds. There is no fire-control API. No kinetic verbs anywhere. Skylock signals. The commander decides. ARGUS chat answers questions like, why did this escalate, with citations to specific signed receipts. Maven AIP Ontology integration is designed. ATAK, Slack, Teams, Bland AI, and Foundry payloads are previewable today and go live by setting one environment variable. Public GitHub repository: https://github.com/Sidra/skylock-hack - **Team:** [Sidra Miconi](https://cerebralvalley.ai/u/Sidra) - **GitHub:** https://github.com/Sidra/skylock-hack - **Demo video:** https://youtu.be/s_5I8BOVrZk - **Project:** https://cerebralvalley.ai/e/3rd-annual-natsec-hackathon/hackathon/gallery?project=19 ### 80. Hedgehog Loop The perception model of a UAV that crushed tank detection yesterday fails the moment the adversary tries to camouflage it: cope cages, camo netting, hedgehog welding. We are using these systems today. Every minute they don't work is a burden on us. Collecting training data from the real world takes weeks or months. The war is evolving faster than that, by then, there will be new modifications. Whose problem is this? The drone operator at the tactical edge, a soldier with a backpack, a controller, and minutes to decide. They have first-hand knowledge of what changed on the battlefield and why it's confusing the autonomy. Today they have no way to push that knowledge back into training without a multi-week data pipeline. The loop we want: the operator sends a note and a few screenshots from the flight controller, and the model is updated overnight. - **Team:** [Roland Pinter](https://cerebralvalley.ai/u/rolandpinter), [Domonkos Haffner](https://cerebralvalley.ai/u/domonkos) - **GitHub:** https://github.com/roland-diffusedrive/natsechackathon - **Demo video:** https://youtu.be/nfDC1gbCtuU - **Project:** https://cerebralvalley.ai/e/3rd-annual-natsec-hackathon/hackathon/gallery?project=20 ### 81. Liminal Liminal Custody, a pre-command evidence integrity system, ensures that the operator remains in the judgment seat rather than the correlation seat. Most command-and-control systems begin after a signal becomes commandable. Liminal Custody handles the preceding layer by ingesting open-source intelligence (sanctions lists, Global Fishing Watch records, Sentinel imagery, and archived AIS data), aggregating it into a knowledge graph, and allowing bounded agents to propose hypotheses. The analyst can click into a hypothesis to review the entire reasoning chain, which consists of individually unremarkable markers chained together to form a pattern of suspicious intent. The agent performs the correlation work, while the operator makes the judgment. The cognitive pipeline operates in a left-to-right manner: Ingest, Relate, Hypothesize, and Review with Reasoning. Liminal Custody was demonstrated on real Strait of Hormuz OSINT data, specifically an Iranian crude tanker (HUGE / IMO 9357183) that has been spoofing MMSI through seven broadcast identities since 2012. OFAC-sanctioned vessels engaged in gray-market evasion, loitering clusters, and ships changing identity over time. Each individual behavior could have gone unnoticed, but when chained together, the system revealed the pattern. Refusal is structurally enforced rather than requested. Every AI specialist output passes through a server-side guard that enforces citation minimums, indicator-evidence requirements, and posterior thresholds. AI cannot overclaim; restraint is a security feature. The track consists of three primary components: PS1 (Sensor Analysis & Integration), PS4 (Digital Defense), and PS3 (Mission C2) architecture. The structural guard in PS4 treats the evidence chain as a defensible attack surface, while PS3 provides a narrative that allows us to slot one layer earlier than CJADC2 or Maven Smart. System. Team: Shruti Rajagopal (founder, Liminal — pre-decisional infrastructure for unresolved context) and Shayaun (Sean) Nejad (co-founder, OffSec-certified offensive-security professional, top-100 Hack The Box; the structural guard is his architecture). Pilot Ask: 90 days with a maritime watchfloor or xTech-aligned sponsor on contested AIS replay data. Maven is the foundation; we are the substrate. - **Team:** [Shruti](https://cerebralvalley.ai/u/LiminalShruti), [Sean Nejad](https://cerebralvalley.ai/u/kuzushi_security) - **GitHub:** https://github.com/liminalshruti/liminal-natsec - **Demo video:** https://www.loom.com/share/5f522ce2ba4c45c8921322d8869833c9 - **Project:** https://cerebralvalley.ai/e/3rd-annual-natsec-hackathon/hackathon/gallery?project=22 ### 82. Barad-dûr Airfields are high-signal locations where subtle visual changes can indicate meaningful shifts in activity, but manually reviewing satellite imagery over time is slow, inconsistent, and easy to miss. This project turns the open-ended question “let me know when there is cause for concern” into a structured surveillance and analysis pipeline for designated locations of interest, starting with airports and airfields. It combines proven computer vision and object detection methods to identify aircraft, model-assisted analysis to classify aircraft type and assess visual reliability, and generative AI to summarize changes, uncertainty, and potential anomalies in analyst-readable language. As AI capabilities advance, their value still depends on the quality of the data and context provided; the goal here is not just to count planes, but to understand how aircraft composition, concentration, and surrounding context evolve over time, then surface unusual changes for timely human intervention - **Team:** [Christine Baek](https://cerebralvalley.ai/u/cbaek) - **GitHub:** https://github.com/viirii/barad_dur - **Demo video:** https://youtu.be/FehKX-S4I00 - **Project:** https://cerebralvalley.ai/e/3rd-annual-natsec-hackathon/hackathon/gallery?project=63 ### 83. Team Osing Montioring Critical infrastructure OSINT monitoring: Specifically, the war with Iran has highlighted how seemingly random tweets and other social media posts have been used for targeting decisions by our adversaries. A dynamically responsive system that would flag such occurances would provide early warning for potential targets. - **Team:** [Srikar Voleti](https://cerebralvalley.ai/u/srikar-voleti) - **GitHub:** https://github.com/vsrikar08/osint-monitoring - **Demo video:** https://youtu.be/RoqacR0kMC0 - **Project:** https://cerebralvalley.ai/e/3rd-annual-natsec-hackathon/hackathon/gallery?project=77 ### 84. TAC-FUSE TAC-FUSE is a local-first edge C2 node for Problem Statement 2: Edge Deployments and Drone Operation. The demo proves that a front-line operator can keep command authority over drones from a hardened laptop or backpack-class kit when central connectivity is intermittent, degraded, or fully denied. The primary capability is not object detection. The primary capability is resilient local C2: operator tasking, drone state, audit log, route/geofence alerts, and deferred sync all persist in local SQLite (`mission.db`) before any external service is contacted. Sensor inference and accelerator paths demonstrate useful local processing once the C2 loop is already working. - **Team:** [Jason Lu](https://cerebralvalley.ai/u/kearm) - **GitHub:** https://github.com/Nottlespike/TAC-FUSE - **Demo video:** https://www.youtube.com/watch?v=j9L3jL1RtNE - **Project:** https://cerebralvalley.ai/e/3rd-annual-natsec-hackathon/hackathon/gallery?project=24 ### 85. TAK Phone We embed TAK into a phone so it survives a factory reset and we can get PLI and sensor data off. - **Placement:** 4th Place - **Team:** [Evin Revello](https://cerebralvalley.ai/u/evrev), [Nathan Fry](https://cerebralvalley.ai/u/profet) - **GitHub:** https://github.com/evinr/tak-phone - **Demo video:** https://www.youtube.com/watch?v=bB0epYma3tU - **Project:** https://cerebralvalley.ai/e/3rd-annual-natsec-hackathon/hackathon/gallery?project=65 ### 86. Fantom Finder Introducing Fantom Finder — passive-radar verification of cooperative broadcasts. $500 of off-the-shelf SDR hardware listens for radar reflections off existing TV broadcast towers and cross-checks every Remote ID drone broadcast, AIS ship transponder, and ADS-B aircraft squitter against what physics actually shows. Mismatches auto-broadcast as Cursor-on-Target XML to ATAK / WinTAK / Delta in sub-millisecond. Use cases in strait of Hormuz and on the battlefield to detect dangerous signals. - **Team:** [Adam Tang](https://cerebralvalley.ai/u/antangled), [David Dominguez Hooper](https://cerebralvalley.ai/u/davidhooper) - **GitHub:** https://github.com/eldaeonuap/phantom-proof - **Demo video:** https://youtu.be/GyclVVk0gwM - **Project:** https://cerebralvalley.ai/e/3rd-annual-natsec-hackathon/hackathon/gallery?project=68 ### 87. BLACK | FIBER FOG drones are immune to jamming. The current answer is barbed wire. BlackFiber neutralizes them with an edge-autonomous Palantir CASK + laser gimbal that breaks the fiber's cladding and drops the control link. - **Team:** [Justin Stefan Stoica-Tica](https://cerebralvalley.ai/u/justinnst), [Volodymyr Silchenko](https://cerebralvalley.ai/u/volod), [Filip Buscu](https://cerebralvalley.ai/u/filipb) - **GitHub:** https://github.com/jusstinn/Katena - **Demo video:** https://youtu.be/Dcbhx9owgbg - **Project:** https://cerebralvalley.ai/e/3rd-annual-natsec-hackathon/hackathon/gallery?project=70 ### 88. Mobius Logic Inc. Our project will address Capability 3: Mission Command and Control. In many cases, the real-time detection and anticipation of other actors’ (opponents' or otherwise) intentions is critical to the success of an actor in multi-actor, goal-directed situations. Automating this detection is particularly important in the world of semi-autonomous systems. In practice, an opponent's intent can often be mapped to a series of decision points, where they may take actions given their perceived likelihood of success. Games like Texas Hold’ Em Poker provide a good testbed for using these decision points to automatically detect an opponent’s intent mid-game, and adjust our own strategy accordingly. We will begin by creating LLM agents with different poker player “personalities” (e.g. a reckless agent that often chooses to go all-in, a cautious agent that folds when they are unlikely to win, etc.). We will then collect observations from these agents and test clustering algorithms like BIRCH, Spectral, and Ward to differentiate the personalities in the observation space. Success from here involves finding the points in the game where the personalities can be differentiated and exploiting them online to force an adversary to reveal their personality. E.g. one decision point may occur immediately after we choose to raise following the flop. Based on our observation classification, we may know that a reckless opponent would choose to go all in here, whereas a cautious agent may choose to call. By choosing to raise at this point, we compel our adversary to choose an action consistent with only one of our personality paradigms, thus giving us insight into its intentions. The methodology we develop and test here will help form the groundwork for decision-making procedure that lets us determine and drive opponent actions in real time. It has many applications to Command and Control, including online intent determination, inverse RL, and threat classification. In addition, it can be used offline to enhance Agent Training and identify gaps in the known space of possible agent behaviors. - **Team:** [Sarah Mansoor](https://cerebralvalley.ai/u/sarahmansoor), [Savana Ammons](https://cerebralvalley.ai/u/sjammons) - **GitHub:** https://github.com/mobius-logic/leduc_holdem_adversaries/tree/main - **Demo video:** https://www.youtube.com/channel/UCA-D4bQWUkFypHg0oEaKevQ - **Project:** https://cerebralvalley.ai/e/3rd-annual-natsec-hackathon/hackathon/gallery?project=71 ### 89. Argus - DS EMCON Sentinel is an open-source ATAK-CIV plugin that protects drone operators by detecting their radio emissions, modeling enemy signal-tracking capabilities, and warning them when they’re at risk of being located and targeted—providing real-time guidance to reduce exposure and stay safe. - **Team:** [Sellab Ahmadzai](https://cerebralvalley.ai/u/Argus-Defense), [Hrittik Chatterjee](https://cerebralvalley.ai/u/Converx), [Akash Karthik](https://cerebralvalley.ai/u/veeha12355) - **GitHub:** https://github.com/PoggyBobby/EMCON-Sentinel - **Demo video:** https://youtu.be/LKeLwimOees - **Project:** https://cerebralvalley.ai/e/3rd-annual-natsec-hackathon/hackathon/gallery?project=76 ### 90. Superman Our project aims to enable J3 (Operations), J5 (Plans/Strategy) and political advisors to make decisions and execute kill chains with the context of political and psychological implications, e.g. the effect on war support from the enemy civilian population, with the help of natural language input for creative orders - **Team:** [Charlie Aram](https://cerebralvalley.ai/u/charliearam), [Sarah Golden](https://cerebralvalley.ai/u/srgjasmine) - **GitHub:** http://github.com/sgolden26/Superman - **Demo video:** https://youtu.be/sHM_vqkxg-I - **Project:** https://cerebralvalley.ai/e/3rd-annual-natsec-hackathon/hackathon/gallery?project=78 ### 91. EdgeLink An ISR drone command system controlled entirely through natural language via Claude Code and a custom MCP server. A single operator can arm, take off, reposition, and assign patrol/orbit behaviors to a live ArduCopter SITL drone in Gazebo — no manual controls. The MCP server exposes drone tasking tools (goto coordinates, assign to named location, search patterns, swarm status) that Claude calls directly. A FastAPI world-state layer ingests telemetry and VLM detections and serves them back to the LLM for situational awareness. - **Team:** [Kyle Orciuch](https://cerebralvalley.ai/u/korciuch), [Althea Hudson](https://cerebralvalley.ai/u/altheah), [Charlie Gordon](https://cerebralvalley.ai/u/csggordon) - **GitHub:** https://github.com/csggordon/cerebralvalley - **Demo video:** https://youtu.be/P3ndcE4xxXo - **Project:** https://cerebralvalley.ai/e/3rd-annual-natsec-hackathon/hackathon/gallery?project=80 ### 92. Dirty Deeds We had to update the link from youtube as it would not generate the link at https://studio.youtube.com/channel/UCiYqw2-65DuCkB0WMsu2nSQ -> USE VIMEO https://vimeo.com/manage/videos/1188853828 Project Dirty Deeds: We are running inside of Palantir for this demonstration. We are not able to take the code outside of the platform. Operator data is never exposed. Nothing is transmitted in a human readable format. You need the corresponding key to un-encrypt any and all data flowing across the chain. Overview of Solution - We built a solution for military and intel use that we believe cross cuts many of the problem statements outlined for this hack. A "Non-transferable, hardware-bound, software-attested identity.” We take cryptography and make it more lightweight. We implemented a SBNFT structure for the attestation layer, implementing Post Quantum Cryptography in an Open Attestation Protocol within a 23 byte message for Digital Defense. We believe, what we bring, also solves problems that exist within Mission Command and Control by improving decision making that accelerates the kill chain while painting a unified, explainable and forensically viable operational picture. Real time attestation - Please refer to video on the Palantir platform. On Chain recording - As you can see, we are attesting on chain in real-time. The table above gives examples of situations where attestations should be made. Post Mortem analysis - There are several reasons we believe this is helpful to our intel and warfighter communities. Often, a soldier is asked to recall details - hardware lapses, signal failures, and so on. Now, the AAR can include these details in a cryptographic sense. How, when or from whom did the delay originate. Is this situational, or systemic? Having this information to review in the AAR, along with every other decision and data point used in the battle space, would take the burden off of the operator and give a new level of actionable data that would inevitably lead to quicker identification of problems and deployment of solutions. We’re going for fast and accurate, explainable and responsible. Conclusion: We believe we can apply this model to everything from tanks to comms links. The entire problematic history of an attested piece of hardware available on a tablet that is authorized and attested. Sensor or radio communications attested as they arrive to a fellow attested device. No man in the middle attack. The operator's SBNFT proves device + person + software. The attestation on the brief proves it was not modified. The chain of custody is cryptographic — not dependent on logs that can be deleted, or screenshots that can be faked.” One audit trail. No vendor silos. GPS Denied offline verification - no network required. Zero code changes. - **Team:** [Silicon Valley](https://cerebralvalley.ai/u/TechExplorGroup) - **GitHub:** https://github.com/aruracode/xTech - **Demo video:** https://vimeo.com/1188853828 - **Project:** https://cerebralvalley.ai/e/3rd-annual-natsec-hackathon/hackathon/gallery?project=87 ### 93. EAGLE WireBird protects critical sites from drone swarm attacks with autonomous drones that perch, recharge, and relaunch from overhead powerlines. Enemy drones are becoming cheaper, faster, and easier to deploy at scale, creating a timing gap in today’s defense model. Ground-based systems are hard to scale, and interceptor drones only work if they are close enough, charged enough, and ready before the swarm arrives. WireBird turns existing overhead powerlines into forward recharge points for defense drones. This removes battery swaps, reduces range dependency, and enables persistent aerial sentries to stay on station near vulnerable sites. Our retrofit kit upgrades existing drones into field-persistent defense assets, enabling faster interception, longer endurance, and protection domes wherever powerlines exist. - **Team:** [Anis cheriet](https://cerebralvalley.ai/u/Anis) - **GitHub:** https://github.com/arminforoughi/eaglecharge - **Demo video:** https://youtu.be/TtLx_mZzY-0?si=FB7S6WMrCly-1ufF - **Project:** https://cerebralvalley.ai/e/3rd-annual-natsec-hackathon/hackathon/gallery?project=89 ### 94. sentinel EW sension fusion for on ground soldiers at the tactical edge - **Team:** [Aili Vaananen](https://cerebralvalley.ai/u/Eyelee) - **GitHub:** https://github.com/samanthalee1103/ew-fusion/ - **Demo video:** https://youtu.be/FeHwIdZzws8 - **Project:** https://cerebralvalley.ai/e/3rd-annual-natsec-hackathon/hackathon/gallery?project=91 ### 95. Team LATTICE Performance‑focused, latency telemetry tool for inferring coarse network location and falsifying claimed regions using UDP echo probes, kernel RX timestamps, split‑probe paths, and a native dashboard with calibration and drift awareness. - **Team:** [Salvador Escobedo](https://cerebralvalley.ai/u/svdr) - **GitHub:** https://github.com/svdrecbd/LATTICE - **Demo video:** https://youtu.be/_XryzGHpq2Y - **Project:** https://cerebralvalley.ai/e/3rd-annual-natsec-hackathon/hackathon/gallery?project=93 ### 96. Vignesh Swarm of robots, autonomous agents, ar for soldiers - **Team:** [Vignesh Anand](https://cerebralvalley.ai/u/machgerita) - **GitHub:** https://github.com/vignesh-anand/mumt/ - **Demo video:** https://youtu.be/ANpD3y4TwFw - **Project:** https://cerebralvalley.ai/e/3rd-annual-natsec-hackathon/hackathon/gallery?project=96 ### 97. CANOPY: Cross-domain Attribution and Orbital Protection sYstem "We must defend U.S. space capabilities, and we must protect our forces from space-enabled attack." — CSO Gen. Saltzman, Space Warfighting, March 2025. Space supports every fight. CANOPY lets every fight support space. A brigade in the field depends on space for everything: GPS, encrypted radio, blue force tracker, satellite imagery. But adversaries no longer attack space alone. They jam signals on the ground, probe networks through cyber, and maneuver inspector satellites in orbit. Today, no one fuses those signals fast enough to see it. CANOPY does. It ingests data from every domain: RF, cyber, OSINT, HUMINT, satellite tracks, GPS integrity, comms health, and drone sensors. It turns this into a single decision a brigade commander can act on. Three AI agents (gemma4:e2b) make that decision auditable: one names the adversary, a red team challenges it, and a reconciler produces the final assessment with calibrated confidence. Every action is grounded in real orbital mechanics (Skyfield/SGP4) and a doctrine-grounded threat knowledge base with ROE. Every step is visible and traceable on screen. The output is one of three actions: passive defense, active defense, or strike request, which are routed automatically to the proper authority. The commander gets one button. The right response gets the right approval at tactical speed. CANOPY runs at the edge on a single Nvidia Jetson with an RTX 3090, and can go from first anomaly sensed to decision in <20 seconds. We also ship an evaluation harness with an adversarial scenario benchmark to measure attribution accuracy and confidence calibration; this serves as a foundation for frontier labs to red-team agentic systems against real & simulated cross-domain campaigns in Space. - **Placement:** 2nd Place - **Team:** [Brian Wu](https://cerebralvalley.ai/u/brianwu), [Jeewoo Choi](https://cerebralvalley.ai/u/jeewoo), [Lillian Sweet](https://cerebralvalley.ai/u/LillianSweet), [Ethan Lai](https://cerebralvalley.ai/u/ethan13) - **GitHub:** https://github.com/787-10/CANOPY - **Demo video:** https://youtu.be/NzDcP5XryC4 - **Project:** https://cerebralvalley.ai/e/3rd-annual-natsec-hackathon/hackathon/gallery?project=25 ### 98. VICTUS Technologies, Inc. A single operator commands a swarm of autonomous drones from a backpack-portable edge kit. Each drone runs a small reasoning LLM on an NVIDIA Jetson Orin, ingests its own video, and acts on terse, doctrine-shaped commands sent from a Palantir Foundry orchestrator — even when the link is intermittent. - **Team:** [Stephen Moskal](https://cerebralvalley.ai/u/SMoskal), [Garrett Harty](https://cerebralvalley.ai/u/gharty-victus), [Jesse Hamel](https://cerebralvalley.ai/u/VICTUS01) - **GitHub:** https://github.com/stephen-moskal/natsec-hackathon-2026-victus - **Demo video:** https://youtu.be/Kz3l-NRidVw - **Project:** https://cerebralvalley.ai/e/3rd-annual-natsec-hackathon/hackathon/gallery?project=66 ### 99. Badlands Badlands is a cyber self-play environment for long-horizon, co-evolving attacker and defender evaluation. It helps mission owners see how model behavior, capability, cost, and mission risk evolve over weeks or months instead of relying on one-off benchmarks. Badlands simulates a mission system with users, identity, applications, files, tickets, services, telemetry, deadlines, and operational disruption. Attacker, defender, and mission-user agents interact inside that stateful world over repeated episodes. The core idea is co-evolution: attackers and defenders adapt through role-visible feedback and role-isolated memory while green/user activity keeps mission pressure alive. This lets teams study how agent behavior changes as context, memory, tokens, latency, test-time compute, and system state accumulate. The problem we solve is continuous cyber capability measurement. As AI systems become more capable and inference time compute becomes more valuable than model weights, mission owners need to know whether cyber risk is rising, whether defenders are improving, and where automation creates new operational harm. Badlands makes that measurable and affordable by running locally on OpenAI-compatible model endpoints, compacting long trajectories when needed, and replaying every score from canonical JSONL evidence. - **Team:** [Jarrod Barnes](https://cerebralvalley.ai/u/jarrodbarnes) - **GitHub:** https://github.com/jbarnes850/badlands - **Demo video:** https://screen.studio/share/fsOnKRgl?state=uploading - **Project:** https://cerebralvalley.ai/e/3rd-annual-natsec-hackathon/hackathon/gallery?project=83 ### 100. AWW (Aim With Words) AWW. Aim with Words. The first targeting system that takes orders in English. You tell it 'fire at the electronics, ignore the laptop' it shows you its reasoning, X's out what it excluded, waits for your approval, and then puts a laser on each target with sub-pixel precision. Watch. - **Placement:** 3rd Place - **Team:** [Jesse Chang](https://cerebralvalley.ai/u/JesseChang), [Jonathan Heewon Yang](https://cerebralvalley.ai/u/jyang27), [Aaren Wong](https://cerebralvalley.ai/u/Aaren) - **GitHub:** https://github.com/acme-robotics/robotracker - **Demo video:** https://youtu.be/EblYAh0-v-U - **Project:** https://cerebralvalley.ai/e/3rd-annual-natsec-hackathon/hackathon/gallery?project=92 ### 101. Gundo Gundo is a maritime sensor allocation watch terminal. Last January, 140 vessels went dark off Long Beach. Gundo triages every dark event by duration, behavior, and flag state, ranks them by threat score, classifies vessel intent, predicts the most likely position using a particle filter with 90% confidence bounds, and recommends which sensor to task to collapse the search area. Built on real AIS data from NOAA, Baysian inference, and entropy-based sensor optimization. - **Team:** [Jacob Quisumbing](https://cerebralvalley.ai/u/jacobq), [Ethan Lam](https://cerebralvalley.ai/u/ethanmlam) - **GitHub:** https://github.com/ethanmlam/gundooo - **Demo video:** https://youtu.be/OCDcDUJsYbs - **Project:** https://cerebralvalley.ai/e/3rd-annual-natsec-hackathon/hackathon/gallery?project=26 ### 102. Tactical defense Project Kamikaze is an edge-deployable early warning system for Shahed/Geran-class UAVs. We trained a custom detection model on a hand-annotated dataset and deployed it behind a FastAPI + Next.js stack supporting image, video, and live webcam detection. High-confidence detections are promoted into Palantir as operational air-threat alerts, turning raw model output into a shared analyst-facing workflow. - **Team:** [Izzy Mansurov](https://cerebralvalley.ai/u/imansurov) - **GitHub:** https://github.com/sapoepsilon/shahed-detector - **Demo video:** https://youtu.be/DFyA2qRBKqk - **Project:** https://cerebralvalley.ai/e/3rd-annual-natsec-hackathon/hackathon/gallery?project=27 --- Markdown version of https://cerebralvalley.ai/e/3rd-annual-natsec-hackathon/hackathon/gallery. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt