A voice agent that answers the phone. Callers dial a Twilio number and talk directly to a GPT model that responds in real time. Latency sits in the range of classic human-to-human calls.
How it works
Twilio picks up the call and opens a WebSocket stream to our backend. The backend forwards the audio bidirectionally to the OpenAI Realtime API and ships transcripts and events to a Next.js dashboard in parallel. A TwiML response configures the stream.
Function calls let the model fetch data or trigger actions in the background, for example checking availability or writing appointments into a calendar. The dashboard lets an operator follow the conversation live and step in.
Why
IVR menus are rigid and break on free speech. A speech model listens, asks back, and closes more cases on the first call. We use the stack as the base for inbound work in insurance and customer service.
Stack
- OpenAI Realtime API
- Twilio Voice & Media Streams
- Next.js Dashboard
- Express WebSocket Server
- TwiML