How kamalesh-suresh-kumar/rec-voice-ai-receptionist works
REC Voice AI Receptionist answers callers’ public institutional-information questions through a SIP/Asterisk call, AudioSocket audio gateway, speech recognition, RAG-grounded response generation, and speech playback. A separate ingestion workflow prepares website and PDF material for semantic retrieval. The README also documents evaluation capabilities and a planned Phase 2 for authenticated student queries and receptionist escalation; Phase 2 is not represented as implemented runtime behavior. Relationships below follow shown calls or README architecture statements; some optional and unsampled integrations are omitted.
Main entry points: rag_service.py RAG orchestrator · audio_gateway.py Audio gateway · pipeline.py Ingestion pipeline · evaluate_voice_latency.py Voice latency evaluation
Components 23
Call handling
- Asterisk
- Audio gateway
REC-Voice-AI-Receptionist/backend/app/telephony/audio_gateway.py
Voice processing
- Speech detection
REC-Voice-AI-Receptionist/backend/app/telephony/audio_gateway.py - Speech recognition
REC-Voice-AI-Receptionist/backend/app/voice/stt_service.py - CUDA setup
REC-Voice-AI-Receptionist/backend/app/voice/cuda_setup.py - Speech synthesis
REC-Voice-AI-Receptionist/backend/app/voice/tts_service.py
Knowledge preparation
- Web extraction
REC-Voice-AI-Receptionist/backend/app/ingestion/web_scraper.py - PDF and OCR extraction
REC-Voice-AI-Receptionist/backend/app/ingestion/pdf_parser.py - Ingestion pipeline
REC-Voice-AI-Receptionist/backend/app/ingestion/pipeline.py - Text chunking
REC-Voice-AI-Receptionist/backend/app/ingestion/chunker.py
Grounded answering
- RAG orchestrator
REC-Voice-AI-Receptionist/backend/app/rag/rag_service.py - Semantic retrieval
REC-Voice-AI-Receptionist/backend/app/rag/retriever.py - Embeddings
REC-Voice-AI-Receptionist/backend/app/rag/embeddings.py - Chroma knowledge store
REC-Voice-AI-Receptionist/backend/app/rag/vector_store.py - Grounded response
REC-Voice-AI-Receptionist/backend/app/rag/llm_service.py
Quality evaluation
- RAG evaluation
REC-Voice-AI-Receptionist/backend/app/evaluation/evaluate_rag.py - ARES evaluation
REC-Voice-AI-Receptionist/backend/app/evaluation/evaluate_ares.py - RAGAS evaluation
REC-Voice-AI-Receptionist/backend/app/evaluation/evaluate_ragas.py - Voice latency evaluation
REC-Voice-AI-Receptionist/backend/app/evaluation/evaluate_voice_latency.py
Other components
- SIP caller
- Ollama / Gemma
- College website content
- Institutional PDFs
Connections 27
- SIP caller → Asterisk: calls via SIP
README.md - Asterisk → Audio gateway: opens AudioSocket
README.md - Audio gateway → Speech detection: detects speech
README.md - Audio gateway → Speech recognition: transcribes audio
REC-Voice-AI-Receptionist/backend/app/telephony/audio_gateway.py - Speech recognition → CUDA setup: configures
REC-Voice-AI-Receptionist/backend/app/voice/stt_service.py - Audio gateway → RAG orchestrator: asks query
REC-Voice-AI-Receptionist/backend/app/telephony/audio_gateway.py - RAG orchestrator → Semantic retrieval: searches
REC-Voice-AI-Receptionist/backend/app/rag/rag_service.py - Semantic retrieval → Embeddings: embeds query
REC-Voice-AI-Receptionist/backend/app/rag/retriever.py - Semantic retrieval → Chroma knowledge store: queries
REC-Voice-AI-Receptionist/backend/app/rag/retriever.py - RAG orchestrator → Grounded response: generates answer
REC-Voice-AI-Receptionist/backend/app/rag/rag_service.py - Grounded response → Ollama / Gemma: requests generation
REC-Voice-AI-Receptionist/backend/app/rag/llm_service.py - Audio gateway → Speech synthesis: synthesizes reply
REC-Voice-AI-Receptionist/backend/app/telephony/audio_gateway.py - Audio gateway → Asterisk: sends audio
README.md - Asterisk → SIP caller: plays response
README.md - College website content → Web extraction: provides content
README.md - Institutional PDFs → PDF and OCR extraction: provides documents
README.md - Ingestion pipeline → PDF and OCR extraction: extracts PDFs
REC-Voice-AI-Receptionist/backend/app/ingestion/pipeline.py - Ingestion pipeline → Text chunking: chunks text
REC-Voice-AI-Receptionist/backend/app/ingestion/pipeline.py - Ingestion pipeline → Embeddings: encodes chunks
REC-Voice-AI-Receptionist/backend/app/ingestion/pipeline.py - Ingestion pipeline → Chroma knowledge store: stores chunks
REC-Voice-AI-Receptionist/backend/app/ingestion/pipeline.py - Web extraction → Ingestion pipeline: supplies web content
README.md - RAG evaluation → RAG orchestrator: evaluates answers
REC-Voice-AI-Receptionist/backend/app/evaluation/evaluate_rag.py - ARES evaluation → RAG orchestrator: evaluates quality
REC-Voice-AI-Receptionist/backend/app/evaluation/evaluate_ares.py - RAGAS evaluation → RAG orchestrator: evaluates quality
REC-Voice-AI-Receptionist/backend/app/evaluation/evaluate_ragas.py - Voice latency evaluation → RAG orchestrator: measures latency
REC-Voice-AI-Receptionist/backend/app/evaluation/evaluate_voice_latency.py - Voice latency evaluation → Speech recognition: measures latency
REC-Voice-AI-Receptionist/backend/app/evaluation/evaluate_voice_latency.py - Voice latency evaluation → Speech synthesis: measures latency
REC-Voice-AI-Receptionist/backend/app/evaluation/evaluate_voice_latency.py