A million restaurants, one shared problem
What happens today
Over a million restaurants in the US. 70% are independent. The phone rings during the dinner rush and nobody can pick up. Every missed call is a lost order, a reservation that goes to the place down the street, a customer who never comes back.
What ansr does
An AI voice host that picks up every call. It takes orders, makes reservations, and answers questions so staff can focus on the guests who are actually there. Owner uploads a menu photo, connects a phone number. We handle the rest.
See it in action
Menu setup, voice ordering, and the dashboard updating live.
Three steps to live
Upload menu
Owner takes a photo of their menu. Paper, chalkboard, whatever. Gemini Pro Vision extracts every item, price, and dietary tag. AI host is configured and live in 60 seconds.
Customer calls
ElevenLabs Conversational AI handles the call. Natural voice, not a phone tree. Takes orders, makes reservations, answers questions. Auto-detects language and responds natively.
Dashboard updates
Call ends. Transcript goes to Gemini Flash for structured extraction. Order, reservation, or inquiry data hits Supabase. Owner's dashboard updates in real-time.
End-to-end system flow
MENU PHOTO → GEMINI PRO VISION → STRUCTURED MENU DATA → AI HOST CONFIGURED CUSTOMER CALL → ELEVENLABS VOICE AI (WEBRTC) → NATURAL CONVERSATION ↓ TRANSCRIPT → GEMINI FLASH LITE → ZOD SCHEMA VALIDATION → STRUCTURED JSON ↓ SUPABASE INSERT → REALTIME BROADCAST → DASHBOARD UPDATES LIVE
What it does
Natural voice AI
Real conversations, not phone trees. Warm, efficient, casual but professional. Powered by ElevenLabs over WebRTC.
60-second onboarding
Upload a menu photo. Gemini Pro Vision extracts items, prices, and dietary tags. AI host is live immediately.
Multi-language
Customer calls in Spanish, agent responds in Spanish. Hindi, Arabic, Mandarin. Auto-detection, native responses.
Order extraction
Items, quantities, modifications, prices, totals, pickup times. All extracted via Gemini with Zod schema validation.
Reservation handling
Name, party size, date/time, special requests, dietary notes. Status workflow: Upcoming, Seated, Late, No-Show.
Real-time dashboard
Supabase Realtime pushes updates instantly. No polling. Orders, reservations, and inquiries appear the moment a call ends.
Inquiry logging
Every question asked and every answer given, captured and displayed. Full transparency into what customers ask.
Receipt confirmation
Thermal-style receipt with items, prices, total, and pickup time. Clean confirmation the order went through.
Status workflow
Mark orders Preparing or Ready. Mark reservations Seated. Time context: "IN 25 MIN" or "12 MIN LATE" on every card.
Built with
Zero to Agent: Google DeepMind x Vercel
Built solo in 7 hours at the Zero to Agent hackathon in NYC, March 21 2026. Pitched live with the dashboard on the projector and customers ordering via voice in real-time. Placed 3rd out of all teams.
