AI receptionist for restaurants Phyton, AI, Backend, react
Budget: $1,500 – $3,000 USD
The system should be able to:
- Answer incoming phone calls (guests calling to book, cancel, or ask questions).
Make outbound calls (e.g. call people from a waitlist when a table becomes available).
- Integrate with the restaurant’s booking system (first with Sevenrooms, later potentially with other systems that have open APIs).
Speak naturally using advanced text-to-speech so guests feel they are talking to a real person.
Log all interactions in our own database so we keep ownership of the data.
How it should work (high-level flow):
A guest calls the restaurant → the call is handled via Twilio (telephony).
The guest’s speech is converted to text via Speech-to-Text (Google STT or Whisper).
GPT (OpenAI) interprets the text and decides what the guest wants (book a table, cancel, waitlist, general questions).
The backend saves the request in our database (Postgres on Azure).
The backend then sends the booking request to Sevenrooms API (or another system in the future).
The confirmation is saved back into our database.
The AI generates a response, which is converted to speech via ElevenLabs (TTS) and played back to the guest.
If the AI cannot answer, it should respond politely and, if needed, forward the call to human staff.
Technical components we expect:
Backend: FastAPI (Python), running on Azure.
Database: Postgres on Azure (stores all customer and booking data).
Telephony: Twilio (for inbound/outbound calls and SMS).
Speech-to-Text: Google Speech-to-Text or Whisper API.
AI engine: GPT-4o-mini (OpenAI).
Text-to-Speech: ElevenLabs (primary) with fallback to Google Cloud TTS.
Dashboard: React or Next.js frontend showing real-time bookings, waitlist, and call logs.
Important design choice:
Even though we start with Sevenrooms, the system should be modular. That means:
All data is always stored in our own database first.
The booking system is just an “adapter” (Sevenrooms first, but later OpenTable, TheFork, etc.).
- Answer incoming phone calls (guests calling to book, cancel, or ask questions).
Make outbound calls (e.g. call people from a waitlist when a table becomes available).
- Integrate with the restaurant’s booking system (first with Sevenrooms, later potentially with other systems that have open APIs).
Speak naturally using advanced text-to-speech so guests feel they are talking to a real person.
Log all interactions in our own database so we keep ownership of the data.
How it should work (high-level flow):
A guest calls the restaurant → the call is handled via Twilio (telephony).
The guest’s speech is converted to text via Speech-to-Text (Google STT or Whisper).
GPT (OpenAI) interprets the text and decides what the guest wants (book a table, cancel, waitlist, general questions).
The backend saves the request in our database (Postgres on Azure).
The backend then sends the booking request to Sevenrooms API (or another system in the future).
The confirmation is saved back into our database.
The AI generates a response, which is converted to speech via ElevenLabs (TTS) and played back to the guest.
If the AI cannot answer, it should respond politely and, if needed, forward the call to human staff.
Technical components we expect:
Backend: FastAPI (Python), running on Azure.
Database: Postgres on Azure (stores all customer and booking data).
Telephony: Twilio (for inbound/outbound calls and SMS).
Speech-to-Text: Google Speech-to-Text or Whisper API.
AI engine: GPT-4o-mini (OpenAI).
Text-to-Speech: ElevenLabs (primary) with fallback to Google Cloud TTS.
Dashboard: React or Next.js frontend showing real-time bookings, waitlist, and call logs.
Important design choice:
Even though we start with Sevenrooms, the system should be modular. That means:
All data is always stored in our own database first.
The booking system is just an “adapter” (Sevenrooms first, but later OpenTable, TheFork, etc.).