ProjectBuild an AI Voice Receptionist: the explanation · Module 12: Answering from clinic documents · part 14 of 23
Part 14 · Module 12: Answering from clinic documents
Module 12: Answering from clinic documents
25 min read·
About 45 minutes · Needs: Module 11 (the agent and its tools), Module 10 (languages and scripts) · Code: backend/app/rag/retriever.py, backend/app/rag/chunking.py, backend/app/rag/ingest.py, backend/app/agent/graph.py
Where this fits
Ingestion fills Postgres with clinic document chunks; retrieval (blue: this module) finds the right ones for each question.
When a caller asks about timings, fees, insurance or test preparation, the answer must come from Medilane's own documents. This module stores the clinic's documents as searchable chunks in Postgres and finds the few passages that answer a question, whether it was asked in English, Hindi or Hinglish. A wrong passage here becomes a confident wrong answer, read aloud.
The rest of this project is yours to keep
This project is bought on its own, once, and stays readable afterwards, including the parts added to it later.