IronMiner is a spatial intelligence platform for construction site management built for UMD x IronSite 2026. Three AI agents — powered by the Claude API and GPU processing via Vast.ai — analyze job site video footage to detect congestion, safety violations, and productivity bottlenecks. The React/TypeScript frontend and FastAPI backend are complemented by an Android headset client for live worker streaming over LiveKit, with persistent data stored in Supabase.
- Video Agent ingests site footage, extracts frames, and runs spatial analysis via GPU (Vast.ai)
- Safety Agent detects PPE non-compliance, zone violations, and blocked corridors
- Productivity Agent scores congestion per zone, identifies trade overlap, and suggests resource reallocation
- Review Mode: upload footage, receive AI briefings, browse zone congestion maps and alert feeds
- Live Mode: monitor real-time camera feeds, communicate with workers, and receive instant alerts via WebSockets
- Android headset client streams live video from workers using LiveKit audio/video
- BIM-style zone overlay visualizes spatial layout directly in the dashboard
- Teams board tracks worker assignments and pool allocation across active sites
- Frontend: React, TypeScript, Vite
- Backend: Python, FastAPI, WebSockets
- AI: Claude API (vision + text), Vast.ai (GPU inference)
- Database: Supabase (PostgreSQL)
- Live Comms: LiveKit
- Mobile: Android (Kotlin)
- Containerization: Docker