Silicon Valley AI Hub Hackathon

Advertisers struggle to optimize campaigns across budgets, targeting, creatives, and engagement. Traditional tools fail to unify campaign data, product info, and customer feedback, leaving fragmented insights and missed ROI opportunities. Solution: Snow Optima AI Snow Optima AI is an industry-agnostic platform that unifies structured and unstructured data to tackle ad optimization. Built with Python & Fast MCP for agent orchestration and Snowflake Cortex services: Cortex Analyst processes structured campaign data, Cortex Search ingests PDFs and reviews, and Cortex Agents integrate insights into a unified strategy. NLP extracts sentiment and relevance, while multimodal AI fuses campaign, product, and customer signals. A Next.js dashboard delivers interactive insights, with a recommendation engine suggesting budget shifts, keyword refinements, and creative adjustments. Automated reporting produces manager-ready summaries, and Docker ensures scalable deployment. Showcased with Amazon Ads

Supply Chains are always at risk. From strikes, tariffs, shortages and natural disasters. The effects of this ripple through the eco system with millions at stake. Plexfort spawns CrewAI agents that process unstructured data and structured data together in Snowflake using custom CortexAI functions to unearth risks and provide fallback alternatives. It's like Flexport, but with our twist on it.

Submit video for ADA compliance analysis and present recommendations on how to resolve them.

Fine-tuning gpt-oss 20b model to efficiently detect users intents to eliminate latency bottleneck. We need intent detection that preserves the accuracy of a bigger model while delivering responses fast enough to keep multi-threaded conversations fluid. Our goal is to maintain high-quality intent classification with minimal response times so the main thread remains free most of the time.

Problem Statement: Generate multimodal data to complement or be in distribution with a users existing corpus of data. LabRat is an extension to that helps young research assistants navigate the complexity of experiments with handwritten notes and drawings. It can read drawings and transform it into code suggestions that can be used in a Snowflake notebook environment. I am a junior at a very small school in SF and I am a part of our one of our first student labs, dedicated to Mathematical Modeling. However, we are understaffed for mentors and given the high number of new students coming in, I wanted to create something that could be helpful for them to understand and develop physics-based simulations. I already wanted to use Snowflake as a hub for centralizing the data of our experiment runs, but since we rely mostly on notes for setting up experiments, I sought to create an agent that can read handwritten notes and turn into educational code suggestions.

We solve the problem of shallow, forgettable networking by creating a Human CRM that makes every interaction authentic, contextual & actionable. It learns uniquely from your relationships

citizen opens app → sees hospital availability → sees estimate of how many other citizens are heading towards a hospital with availability → citizen chooses one → receives safe routes assess hospital capacity and environmental conditions using a combination of different data sources (or "modalities"), including satellite imagery analysis, environmental sensor data, and temporal pattern analysis. demonstrates the fusion of object detection, environmental analysis (spectral data), and temporal trend-spotting with a critical cross-validation step.

This is an automated Intern Overlay where user can drag the spawned overlay somewhere on their screen could be a chat, Twitter Feed, News Feed, Stock Market, and when certain actions or trigger words are detected with OCR or the use of AI and image recognition, this will alert the user to the user preference whether it would be text or email.

Dashboards and models often fail silently when data drifts: null spikes, schema changes, or category shifts make KPIs unreliable. Traditional drift tools only flag what changed, leaving teams guessing why. The Multimodal Data Drift Detector combines statistical drift detection with context ingestion from release notes and dashboard screenshots. It detects null %, numeric drift (KS/mean shift), categorical drift (PSI), and outliers, then ingests PDFs and screenshots to capture signals like “currency switched USD→EUR” or “API v2 deployed.” The system generates a plain-English narrative with actionable next steps such as validating ETL, backfilling data, or updating contracts. A simulated chat integration enables instant alerts and team responses. An optional sandbox supports threshold tuning, cost-sensitive metrics, and reproducibility with downloadable configs. This helps teams move beyond detection to diagnosis and action, reducing downtime and keeping metrics trustworthy.

YouTube Voice Control Agent The agent helps people with motor disabilities and enables hands free control in general. The agent operates in two modes: Training Mode: Learns your unique voice signature for secure access. Control Mode: Listens for the wake word ("YouTube"), lowers volume, captures your command, matches your voice, and controls playback: Supports multiple commands - jump to , rewind, backward, faster, slower, , search, mute, unmute, pause, resume and more. Social Impact: Enables independent, hands-free video control for users with motor challenges using personalized voice commands. No data leaves your system. All processing is done locally with Python and browser automation for privacy and latency.

⏺ SafAI is an intelligent Gmail monitoring system that provides enterprise-grade security through AI-powered threat detection. The application continuously monitors Gmail inboxes using real-time IMAP connections, automatically analyzing incoming emails for potential security threats including malicious attachments, suspicious images, and dangerous content. Built with a hybrid AI architecture combining Anthropic's Claude for comprehensive text analysis and Google's Gemini Nano-Banana for specialized image annotation, SafAI delivers accurate threat assessment with visual evidence. The system features a React-based dashboard for real-time notifications, detailed threat reports, and annotated security findings, making it an essential tool for organizations seeking proactive email security monitoring and incident response capabilities.

This project is a full-stack web application that lets users upload photos of their shopping receipts and instantly see their spending summarized and categorized. An AI agent—built using Large Language Models (LLMs) and Snowflake’s AI features—automatically reads each receipt, identifies the purchased items, classifies them into spending categories (e.g., Groceries, Household, Beverages, Personal Care), and updates a real-time dashboard of spending insights.

SynthSight uses a knowledge graph to map relationships between data, policies, and compliance rules, then leverages LLMs to make intelligent decisions about data governance. The system analyzes patterns in your data, connects them to regulatory requirements through the knowledge graph, and uses AI reasoning to determine if data is compliant, flag violations, and suggest corrective actions automatically.

Our ocean voyage simulator creates a relaxing and enlightening virtual vacation for you. Tell it where you want to go and the route planner will have you on your way: identifying points of interest, sampling the real weather and wave conditions for your trip, and narrating the whole journey in a pleasing, soothing voice.

Touchdown is where long videos turn into instant highlights. No more scrubbing through hours — our AI captures the moments that matter most and turns them into reels you can watch and share. Starting with football, which can be scaled across every sport, every event, every video. From the stadium to the classroom to your favorite shows — Touchdown makes highlights personalized, automatic, and endless.

This is a tool that helps give emotional clarity and advice actively in conversation with phone calls

Designed for busy parents who might not be able to carry bedtime story books when they travel/facetime with their kids. V1 is integrated with Writer.ai's API and the next iteration will integrate Elevenlab's voice API to generate the parents' voiceover.

MCP agent for housing property management The project uses 2 GitHub repo: https://github.com/williamQyq/Housing-Property-Agent-MCP.git https://github.com/williamQyq/Housing-Agent-Lovable.git

RescueLens is an AI assisted triage agenet. in disaster situations where first responders cant physically assess affected people, this ai agent can be incorporated into drones/robots. It gives an immediate triage based on video pattersn and audio (shouts of help/Im fine) and assesses. There is an override functionality if the doctor believes the assessment is incorrect

Waste Management AI

PawLens turns any smartphone video into instant dog behavior analysis using AI — bringing $200-per-hour veterinary expertise to every pet owner. By analyzing video and audio, it detects aggression, anxiety, and energy levels, highlights key moments with interactive playback, and even offers breed insights and personalized care tips. Think of it as a real-time dog translator that helps 90 million owners prevent problems, keep walks safe, and understand their dogs like never before.

I built a technical AI co-founder, powered by AWS, CrewAI, and Snowflake, that reads your business documents, tracks meetings, analyzes investor and market data, and gives you smart, actionable advice for founders, CEOs, HR managers, and busy professionals all in one place.

MediShare is a simplified medical sharing platform that enables patients to upload documents, generate medical timelines, and securely share their health data with doctors who can then analyze it using AI-powered chat functionality.

Generates memes based on trending searchs on Google. Demo video n/a

multimodal solution to play with medical data