Built a full-stack AI phone ordering platform from scratch for
independent restaurants in Australia. When a customer calls the
restaurant’s number, an AI assistant named Maya answers, takes
the food order conversationally, and sends it live to a kitchen
tablet dashboard — no staff needed to answer the phone.
What the system does:
- AI agent handles the full order conversation — menu,
modifiers, customer name, pickup time, order confirmation - Orders appear in real time on a kitchen tablet (kanban board:
New → In Kitchen → Ready for Pickup) - Complex calls transfer live to the owner’s mobile via Twilio
- One AI agent serves unlimited restaurants dynamically — the
correct menu and config loads per call based on which number
was dialled - Restaurant owners manage hours, menu, and settings via a
web dashboard
Technical highlights:
- Pre-warm cache loads menu data during the ring — eliminating
startup silence before the agent speaks - Server-side variable resolution for Retell AI’s tool calling
(platform limitation workaround) - Timezone-aware trading hours engine with special day overrides
- Supabase Realtime subscriptions with polling fallback for
the kitchen tablet - Role-based auth — admin and kitchen staff roles with
separate access levels
Stack: Retell AI · GPT-4.1-mini · Twilio · Node.js · Express ·
TypeScript · Supabase · PostgreSQL · Next.js 14 · Tailwind CSS ·
Railway · Vercel