AI voice receptionist for Flance

Callie.

The AI receptionist that answers your phone.

Callie holds a real conversation over a real phone line — in real time, with a real voice, and no script running out from under her.

Call Callie +1 (XXX) XXX-XXXX Live demo — real AI, real phone line
What she does on the call

A receptionist, not a phone tree

Callie carries the conversation herself. No menus, no "press 1," no dead air while something loads.

Answers instantly

Service questions get handled live, in a natural voice, the moment the caller asks — no hold music, no waiting for a human.

Books callbacks

When something needs a human, she schedules the callback herself — so no lead sits waiting on hold to go cold.

Files support tickets

Every issue becomes a structured ticket — tagged and routed to the right queue before your team ever picks up.

Takes interruptions in stride

Talk over her mid-sentence and she stops, listens, and picks the thread back up — the way a person would.

Sends a summary after every call

Intent, contact details, next steps — written up and waiting in your system the moment she hangs up.

Under the surface

How a call actually moves

One path is latency-sensitive and stays entirely in-memory over WebSockets. The other runs after the call ends and can afford to queue.

Real-time voice path
  1. Caller
    dials the number
  2. Twilio
    Media Streams · WebSocket
  3. FastAPI
    manages the live session
  4. Gemini Live
    reasoning + voice, ElevenLabs assist

↩ audio streams straight back to the caller — same connection, no round trip to a queue

Post-call pipeline
  1. Status callback
    Twilio webhook, verified
  2. Celery
    broker: Redis
  3. PostgreSQL
    durable, structured storage
  4. Ticket
    intent, contact, next step
Under the hood

The stack, by role

Voice pipeline

Twilio Media Streams FastAPI Gemini Live API ElevenLabs

Data & queue

PostgreSQL Redis Celery

Infrastructure

Docker