# VideoLab

- **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:** [Balamurugan Ramakrishnan](https://cerebralvalley.ai/u/Balamurugan)
- **GitHub:** https://github.com/Balu-SG/videolab
- **Demo video:** https://youtu.be/MIljAJWmfnU
- **Gallery:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/87

VideoLab turns a phone into a physics measurement instrument. Film a pendulum, a falling ball or a spring in 240 fps slow motion with a plain A4 sheet on the wall, and VideoLab returns real physical constants (g, period, spring constant k) with honest error bars.

SAM 3.1 tracks the object through every frame from a single tap. The A4 sheet gives a homography that turns pixels into metres and corrects camera tilt. Deterministic physics fits (damped amplitude-corrected pendulum, free-fall parabola with automatic release/impact, damped spring oscillation) produce the numbers, never the LLM. A Muse Spark agent with five tools plans the setup, runs quality checks and explains the result in a predict → observe → explain loop. Every number it writes is grounded in tool output and is a clickable chip that jumps the video to the frames behind it.

Problem: millions of students learn physics without a lab. Accuracy: 20 benchmark runs with exact ground truth (tilted/far camera, dim light, noise, motion blur) all land within 2σ. Spring k within 0.03%, pendulum g within 0.02% with the string length entered, free-fall g within 1.3%.

Live: https://balmca28--videolab-web.us-east.modal.run
Accuracy report: https://github.com/Balu-SG/videolab/blob/main/docs/ACCURACY.md

## More from Meta Global AI Developer Hackathon

- [BARRED](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/83)
- [ContextPane](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/85)
- [Volgat](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/86)
- [Doodytoon](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/88)
- [Arabian Ai Camp](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/89)
- [Listing Pilot](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/90)

---

Markdown version of https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/87. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt