Product demo
Tableside, start to finish, in about a minute.
An AI waiter for restaurants that I designed and built on my own. A guest with a dairy allergy asks what they can eat, orders by chatting, and the order reaches the kitchen.
Recorded live on a demo restaurant with made-up data. The AI replies are real; the waits between them are cut, the kitchen section is sped up, and the narration is a computer voice.
What to watch for
It answers from real data
The waiter checks the restaurant's own allergen records, separates what the restaurant has marked from what it hasn't, and tells guests to check with staff.
It can't act alone
The model can only call tools. Prices are checked against tool results, and a code gate refuses any order the guest hasn't confirmed.
Staff get an assistant too
Kitchen staff ask how service is going, or take a dish off the menu in a plain sentence, with an undo.
How it's built
TypeScript, Next.js, React and PostgreSQL on Supabase. An LLM agent with 17 schema-validated tools, reached through OpenRouter, with model fallback, guardrails and an evaluation harness that replays scripted guest conversations through the real agent.