Clinical RAG Pipeline with Source Citations & Alerts
Overview
A Retrieval-Augmented Generation pipeline that reads a visitor's description of their own concern and returns the procedures the clinic actually offers for it, every answer cited back to the document it came from. The data was provided by the client in the form of supporting PDFs and EXCEL files for details of procedures that can be performed in their aesthetic clinic for the specific issue that the users pre-diagnose. This app was just to give users an idea of what they can expect to be performed in the clinic. A final diagnosis was always provided by professionals.
Outcome
A patient-facing guidance tool that turns a described concern into the clinic's own relevant procedures, with citations to the source document and contraindication alerts. It sets expectations before the visit — the final diagnosis always came from the clinic's professionals.
Screens & Demos
Architecture & Pipeline
-
01
Procedure PDFs & Excel
Procedure catalogue
- Excel
-
02
Document Parser
Python · procedure extraction
- Python
-
03
Vector Index
BioBERT embeddings
- BioBERT
-
04
Visitor's Concern
Free text · self pre-diagnosis
-
05
Hybrid Search
LangChain · matched to catalogue
- LangChain
-
06
LLM
Llama / Mistral
- LLM
- Llama
- Mistral
-
07
Safety Filters
Contraindication alerts
-
08
Cited Response
FastAPI · links to source document
- FastAPI
End-to-end flow, in order. Scroll sideways on a narrow screen.
Key Features
- Query processor that maps a visitor's described concern to procedure categories
- Vector index built from the client's own procedure PDFs and Excel sheets
- Hybrid search RAG pipeline over the clinic's catalogue
- Every answer cited back to the source document it came from
- Secure API with safety filters and contraindication alerts
- Optional demo interface