# Artiom Varvarenko

- **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:** [Artiom Varvarenko](https://cerebralvalley.ai/u/artiomvarvarenko)
- **GitHub:** https://github.com/artiom-varvarenko/lightwork
- **Demo video:** https://www.youtube.com/watch?v=5hijfLAvvDw
- **Gallery:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4518

Lightwork turns any table into a helper. A phone projector draws on the table, a webcam looks down from above, and your hands are the controls.

You hold a fingertip on a projected mic bubble and say what you want: "help me make pancakes with this", "help me build this Lego", "what is this book about?", "show me a cargo ship", "take me to Paris", or "let's play chess". The table goes dark for a moment so the camera gets a clean picture. Meta's models then work out what is on the table, ask only what a camera can't know (it counts your eggs but asks how many grams are in the bowl), and guide you step by step with light on the real objects: what to take is outlined, an arrow shows where it goes, and the camera checks each step before moving on.

The problem it solves: instructions live on screens, but the work happens on the table. Recipes, manuals and tutorials make you look away, keep your place and judge for yourself whether you did it right. Lightwork puts the instruction where your hands already are and checks it, so you never look away from the work, and there is no screen to touch and nothing to type. It also answers questions about books on the table with a spoken answer and a projected diagram, builds explorable 3D models you can turn, slice open and move through with your hands, flies you through 3D cities, and plays games with you, including chess with your real pieces.

It takes Meta's models off the screen and into the physical world, with a closed loop: the model checks with its own eyes that each real step was done. The hardware is a phone projector, a webcam and a laptop. I built it solo during the hackathon.

## More from Meta Global AI Developer Hackathon

- [Onasafari](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4515)
- [Spark Maxxing](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4516)
- [Aylon Ben-Ami](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4517)
- [Auren](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4519)
- [Spotnack](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4520)
- [deed](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4521)

---

Markdown version of https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/4518. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt