# RestaurantOS 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:** [Vamsi Krishna Reddy](https://cerebralvalley.ai/u/vamsikrishna)
- **GitHub:** https://github.com/vamsikrishna2421/restaurantos-ai
- **Demo video:** https://github.com/vamsikrishna2421/restaurantos-ai/blob/demo-film/film/restaurantos-ai-demo.mp4
- **Gallery:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery
- **Page:** https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1083

Live app: https://restaurantos-ai.fly.dev — tap Try the live demo: order as a Customer (voice ordering in EN/TE/TA/ES) or sign in as Staff.

RestaurantOS AI is an AI general manager for independent, multilingual restaurants.

Guests order the way they talk — in English, Telugu, Tamil, or Spanish — at a kiosk or by phone. Muse Voice Transcribe hears them; Muse Spark turns each utterance into orders over a closed menu, and every station sees only its own dishes, in its own language, with the guest's name on the ticket.

The manager doesn't watch dashboards — the system watches everything else. It forecasts ingredient run-outs from the restaurant's own order history using statistics (zero AI tokens spent watching), and when a threshold trips, Muse Spark narrates a caution with its decision attached: "Paneer runs out about 7:20 tonight. Order now, or take it off the menu." Every caution shows its work — why it fired, which dish is burning the ingredient, which station is a cook short. Nothing acts on its own: the system proposes, a manager approves, and every decision lands in an audit log.

It onboards from a single menu link (AI drafts menu, stations, and roles for one review screen), runs inventory from a counted opening stock — every order consumes ingredients automatically, a shelf photo counts what's there, a supplier bill scan becomes a stock-in draft — and it knows its regulars: when one drifts away it drafts a personal win-back message with their usual on the house, for a manager's approval.

Two protections are built in, not bolted on. AI-drafted ingredient lists stay drafts until the kitchen approves them, and the recipe itself — the process — is never stored, only ingredients and quantities. And for back-office reasoning, ingredient and dish names reach the external model only as per-restaurant codes (a different code in every kitchen), decoded locally; the AI sees codes, not names. Orders from outside systems (any POS or online channel) flow into the same ledger through a source-tagged, idempotent ingestion pipe, so stock and analytics cover every order.

A founder console watches the whole fleet for owners with more than one location.

## More from Meta Global AI Developer Hackathon

- [ASSERTARA](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1080)
- [GhostRep](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1081)
- [Lesstruct](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1082)
- [BenchMate](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1084)
- [Velohud](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1085)
- [Vouch](https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1086)

---

Markdown version of https://cerebralvalley.ai/e/meta-ai-developer-hackathon/hackathon/gallery/1083. Site index for agents: https://cerebralvalley.ai/llms.txt · full text: https://cerebralvalley.ai/llms-full.txt