Case study · AI integration
AI Real Estate Voice Agent
A real-time voice assistant that handles property enquiries: it streams call audio over WebSockets, transcribes speech with Deepgram, and triggers domain-specific actions inside the conversation.
What it does
The agent answers property enquiries in a live voice conversation instead of routing every caller to a person. It listens to the call as it happens, understands what the caller is asking for, and triggers the specific action that enquiry requires.
Implementation
- Twilio WebSockets for real-time call audio streaming
- Deepgram speech-to-text for live transcription of the conversation
- A Python WebSocket server processing live audio streams and dispatching domain-specific function calls
- Modular integration architecture so additional property automation workflows can be added independently
- Ngrok used to expose and test the WebSocket endpoint during development
Engineering highlights
- Real-time constraints — the pipeline is built around streaming rather than request/response
- Conversation logic kept separate from integration concerns, so providers and workflows can change independently
- Function-call dispatch designed as the extension point for new property automations
Try it
A short demo recording walks through the voice agent handling a property enquiry.