# VisionGuard AI

- **Event:** [Meta Global AI Developer Hackathon](https://cerebralvalley.ai/e/meta-ai-developer-hackathon)
- **When:** Oct 26, 12:00 PM – Nov 4, 9:00 PM (EST)
- **Where:** Online
- **Team:** [El Moatasem Madani](https://cerebralvalley.ai/u/El-Moatasem)
- **GitHub:** https://github.com/El-Moatasem/visionguard-ai
- **Demo video:** https://www.youtube.com/watch?v=p-78-WtU5CY
- **Gallery:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1339

VisionGuard AI turns short workplace videos into explainable, human-reviewable safety evidence instead of black-box safety alerts.

A reviewer uploads a video and draws a monitoring zone. Meta SAM 3.1 tracks selected objects, including people, forklifts, cars, shopping carts, and machinery. Muse Spark optionally translates plain-English safety requirements into constrained rule drafts requiring human approval. A deterministic engine evaluates observed conditions and produces potential rule matches with frame ranges, track IDs, and exportable evidence snapshots.

Its core principle is honest uncertainty. When tracking observations are missing, such as during occlusion, VisionGuard does not invent positions or assume safety. Instead, detected gaps become review candidates. Potential forklift operators are identified conservatively for human confirmation rather than automatically classified as pedestrian hazards.

Reviewers can confirm, reject, or leave findings unresolved. Evidence snapshots preserve source hashes, rule parameters, observed frames, and appended review decisions.

VisionGuard has demonstrated real Meta SAM 3.1 tracking on warehouse footage. It also supports retail, parking, and construction workflows through configurable concepts, rules, and six labeled synthetic replay scenarios that run offline without model charges.

VisionGuard AI is a research prototype, not a certified real-time workplace safety monitoring system.

## More from Meta Global AI Developer Hackathon

- [Marketing Studio XR](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1336)
- [Bezenith](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1337)
- [MyDearDaddy](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1338)
- [Ryan Kuederle](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1340)
- [Blackcats](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1341)
- [StarTek](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1342)

---

Markdown version of https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1339. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt