PROJECTS

Production real-time voice conversational assistant built on Pipecat. Features Deepgram Nova-3 STT, emotion-aware Chatterbox TTS (via Resemble AI), hybrid emotion detection (audio + text sentiment), LightRAG graph-based knowledge retrieval, and dynamic A2UI visual responses. Supports barge-in and WebSocket audio streaming with 1-1.5s response times. Connects via Pipecat protobuf WebSocket.