Multi-Agent AI Court for bodycam (3rd place, Milan AI Week 2026)
Built in 7 days solo for the AI Agent Olympics Hackathon — the official hackathon of AI Week 2026 in Milan (the largest AI event in Europe). 3rd place out of 731 teams and 2382 participants.
Problem: Every year, tens of millions of police interactions are recorded on bodycams in the US and EU. Less than 1% ever undergo an audit. When a complaint comes in months later, the officer's written report de facto becomes the sole source of truth. SENTINEL is an audit layer that has never existed: an adversarial multi-agent "court" that processes each recording, cites specific statements and specific frames, and issues a court-admissible verdict in minutes, not months.
Architecture:
Three layers with different speeds and depths of analysis:
Layer 1 — Rapid (
Problem: Every year, tens of millions of police interactions are recorded on bodycams in the US and EU. Less than 1% ever undergo an audit. When a complaint comes in months later, the officer's written report de facto becomes the sole source of truth. SENTINEL is an audit layer that has never existed: an adversarial multi-agent "court" that processes each recording, cites specific statements and specific frames, and issues a court-admissible verdict in minutes, not months.
Architecture:
Three layers with different speeds and depths of analysis:
Layer 1 — Rapid (
Kyiv 
Available for hire