How codesbysaravana/nephele_production works
Nephele is a real-time voice companion for college placement preparation. A student starts a conversation in the frontend, which captures microphone audio and streams it to the backend; the backend transcribes speech, generates a streamed answer, synthesizes speech, and returns transcript/audio events for display and playback. The graph includes the documented provider stages and health endpoint. Provider orchestration details are partly documented in the README; unsampled frontend entry-point wiring is included only where documented.
Main entry points: voice_routes.py Voice WebSocket · VoiceSession.ts Voice session · VoiceOverlay.tsx Voice overlay · main.py FastAPI app
Components 13
Voice frontend
- React app
frontend/src/App.tsx - Voice overlay
frontend/src/components/VoiceOverlay.tsx - Voice session
frontend/src/audio/VoiceSession.ts - Voice URL config
frontend/src/config.ts - PCM capture worklet
frontend/public/pcm-processor.js - Audio playback
frontend/src/audio/Playback.ts
Voice backend
- FastAPI app
backend/main.py - Voice WebSocket
backend/routes/voice_routes.py - Health endpoint
backend/main.py
AI providers
- AssemblyAI STT
- OpenAI LLM
- Deepgram TTS
Other components
- Student
Connections 16
- Student → Voice overlay: starts conversation
frontend/src/components/VoiceOverlay.tsx - React app → Voice overlay: renders
README.md - Voice overlay → Voice session: starts session
frontend/src/components/VoiceOverlay.tsx - Voice session → Voice URL config: resolves URL
frontend/src/audio/VoiceSession.ts - Voice session → PCM capture worklet: loads capture
frontend/src/audio/VoiceSession.ts - Voice session → Audio playback: plays audio
frontend/src/audio/VoiceSession.ts - Voice session → Voice WebSocket: streams audio and receives events
README.md - FastAPI app → Voice WebSocket: registers router
backend/main.py - Voice WebSocket → AssemblyAI STT: forwards audio
README.md - AssemblyAI STT → Voice WebSocket: returns transcript
README.md - Voice WebSocket → OpenAI LLM: sends transcript and history
README.md - OpenAI LLM → Voice WebSocket: streams tokens
README.md - Voice WebSocket → Deepgram TTS: sends sentences
backend/routes/voice_routes.py - Deepgram TTS → Voice WebSocket: returns audio
README.md - Voice WebSocket → Voice session: streams audio and transcript events
README.md - FastAPI app → Health endpoint: serves
backend/main.py