LLM · Voice · AI Assistant
BadarAI Assistant
Conversational Voice AI
FastAPIGemini APIDeepgramElevenLabsWebSocketsRedisReactPython
System architecture
- User speechmicrophone
- Streaming STTAIDeepgram
- MemoryRedis
- ReasoningAIGemini · multi-turn memory
- Live toolsreal-time data fetch
- Voice synthesisAIElevenLabs
- Spoken replystreamed audio<800ms
Measured in production
<800msVoice round trip
3Streaming stages: STT, LLM, TTS
Overview
Full-duplex voice AI assistant with streaming STT, LLM reasoning, and TTS synthesis. Deepgram streams speech-to-text in real time; Gemini handles multi-turn dialogue with persistent memory; ElevenLabs renders natural-voice responses under 800ms. Supports persona customisation and live tool-calling for real-time data fetching.
Result
- <800ms end-to-end latency
- Multi-turn memory
- Live tool-calling support