Kitchen tablet. Every order lands here the same way, whether it came from the app, a QR code or a phone call. Tickets show whether an order is paid, never how.
Kitchen tablet. Every order lands here the same way, whether it came from the app, a QR code or a phone call. Tickets show whether an order is paid, never how.
The production app runs on Cloud Run and Cloud SQL, with Gemini for ordering conversations, Cloud Speech-to-Text and Text-to-Speech for voice, and Google Maps Platform for discovery.
This page is a self-contained browser demo. Its ordering logic is a lightweight keyword matcher standing in for Gemini, so it works offline.
There are an estimated 45,000 Chinese restaurants in the US. Many are independent, family-run businesses where phone orders and language barriers are everyday problems.
| Part of the phone line | Today's prototype | What we want to evaluate |
|---|---|---|
| Telephony | Twilio | Twilio, or Google's contact-center stack |
| Voice | Amazon Polly voices (via Twilio) | Gemini real-time voice on Vertex AI |
| Understanding | Gemini 2.0 Flash, one turn at a time | Gemini Live API, interruptible conversation |
| Measured on | Order accuracy against the kitchen ticket · response latency · Chinese and Spanish comprehension · cost per completed order (results to be measured) | |