# witlabs

- **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:** [Prabal Singh](https://cerebralvalley.ai/u/0xPrabal)
- **GitHub:** https://github.com/PrabalParihar/reel2real
- **Demo video:** https://web-production-d1454.up.railway.app/submission/index.html
- **Gallery:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/5311

Reel2Real — From tutorial to tabletop, one step at a time.

Following a hands-on tutorial often means pausing, scrubbing and looking between the lesson and your workspace. Reel2Real is a browser guide for short, safe tabletop lessons with a phone on a stand or a laptop camera.

Upload a tutorial you own or have permission to use, and Muse Spark creates a structured step plan from the complete accepted clip. Alternatively, start with a goal and one workspace still. In the coach, SAM 3.1 locates the current step’s objects in camera keyframes. Learners can adjust framing, request a fresh check and press Next to confirm their own progress. The recap distinguishes learner confirmations, object visibility, skipped steps and unvisited steps.

Built with Next.js, React, TypeScript and Node 22 on Railway, the prototype uses server-side model calls, media validation, bounded requests, cancellation and explicit recovery. Live, Cached and recorded Sample results remain distinct. Five credited recorded lessons let judges explore without a camera or live model calls.

Live goal planning, complete-clip planning and controlled object localization have been demonstrated. Detection still misses some visible objects; physical-phone validation, useful physical-task correction and learner evaluation remain open. Object visibility does not verify actions, technique, order or safety, and no measured learning gains are claimed.

Live app: https://web-production-d1454.up.railway.app/
Narrated prototype demo: https://web-production-d1454.up.railway.app/submission/index.html
The 2:03 demo labels recorded Samples, generated QA media, virtual-camera input, separate live runs and a shortened planning wait.

## More from Meta Global AI Developer Hackathon

- [ULTRAGENIS](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/5308)
- [BYTEWISE](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/5309)
- [Disaster Mitigation Bench](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/5310)
- [FixVision](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/5312)
- [Avatodo](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/5313)
- [Shreynsh Singh](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/5314)

---

Markdown version of https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/5311. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt