Realtime Voice Agent
Low-latency conversational AI with tool execution
Engineered a bidirectional voice agent that achieves sub-200ms response times while executing complex tool calls in real-time, streaming audio directly through WebRTC with live function dispatch.
- The Challenge
- Sub-200ms response times with function calling
- Gemini Live API · WebRTC · FastAPI
- Engineering
- Bidirectional audio streaming with live tool dispatch
- github.com/david-mamani/voice-agent
- Impact
- 150ms avg latency — real-time conversation quality
- Demo at Google DevFest 2025