# EdgeCases

- **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:** [Indranil Palit](https://cerebralvalley.ai/u/indranil_palit)
- **GitHub:** https://github.com/IndranilPalit24/AskFootage
- **Demo video:** https://youtu.be/KFjKkqQqf3o
- **Gallery:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/723

AskFootage lets you ask a plain-English question about recorded camera footage and get back the matching clip plus the proof. Hours of fixed-camera footage are effectively unsearchable today: finding one moment means scrubbing through recordings by hand. With AskFootage you type something like "person carrying a box" and get the clip, the time range, a red box that follows the object on the playing video, and a "Why this matched" panel with the measurements behind the match (the relation, how long it was seen, frames together, motion agreement).

How it works: SAM 3.1, through the Meta Model API, runs once per object type to detect and track objects. The tracks are stored in a local SQLite index. Relations such as "carrying" are resolved by ordinary, testable Python code, so every answer is reproducible and explainable. After indexing, searches are free. Spend is capped in code (hard limit, default $25 per run). There is no face recognition and no identity feature, by design.

The app has a React web UI and a FastAPI backend with a contract-first OpenAPI API. Anyone can run it with no API key: the recorded SAM replies are included and replayed offline. The two real clips are short test videos filmed by the author; the cameras and recordings around them are a disclosed, generated demo dataset.

## More from Meta Global AI Developer Hackathon

- [Avasara Lab](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/720)
- [VANTALab](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/721)
- [Field from Above](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/722)
- [RAJAKUMARAN](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/724)
- [Point, Say, Press](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/725)
- [Agentic Verification Layer](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/726)

---

Markdown version of https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/723. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt