Med Insight AI

Privacy-first medical document analysis with RAG (LangChain + ChromaDB). Upload lab PDFs, ask questions, and get answers with cited sources from a medical dictionary (NIH/Mayo style).

Features

• Upload medical lab PDFs – Native text extraction with OCR fallback for scanned pages
• PII redaction – Names and phone numbers redacted before processing (upload text and query)
• RAG over medical dictionary – Answers grounded in an ingested medical knowledge base
• Source section – Every response includes a Source section citing the medical database
• Simplified vs Technical view – Toggle result display for plain-language or technical context
• WCAG-friendly, calming UI – Accessible, clean, privacy-first frontend

Stack

• Frontend: React 18 + Vite + TypeScript
• Backend: FastAPI (Python)
• RAG: LangChain + ChromaDB; embeddings: sentence-transformers (all-MiniLM-L6-v2)
• OCR: pdf2image + Tesseract (optional; install Tesseract on the host)
• PII: Presidio (optional) + regex for phones and common lab patterns

Quick start

1. Backend

cd backend
python -m venv .venv
source .venv/bin/activate   # or .venv\Scripts\activate on Windows
pip install -r requirements.txt

Install Tesseract for OCR (scanned PDFs):

• macOS: brew install tesseract
• Ubuntu: sudo apt install tesseract-ocr

Ingest the medical dictionary into ChromaDB (run once):

python -m scripts.ingest_medical_dict

Start the API:

uvicorn app.main:app --reload --host 0.0.0.0 --port 8000

2. Frontend

cd frontend
npm install
npm run dev

Open http://localhost:5173. Use the proxy in vite.config.ts so /api goes to the backend.

3. Optional: LLM for answers

By default, the app returns retrieved context and sources without an LLM. To get natural-language answers:

• Install Ollama and run e.g. ollama run llama3.2, or
• Change app/services/rag_service.py to use OpenAI or another LLM.

Project layout

backend/
  app/
    api/          # upload, query, health
    core/         # config
    middleware/   # PII redaction (optional middleware)
    services/    # pii_redaction, pdf_ocr, vector_store, rag_service
  data/           # medical_dictionary_sample.json
  scripts/        # ingest_medical_dict.py
frontend/
  src/
    components/  # UploadSection, QuerySection, ResultSection

Privacy

• Uploaded PDFs are not stored on disk; only redacted extracted text is chunked and embedded.
• Query text is redacted for PII before being sent to the RAG pipeline.
• Use Presidio for stronger NER-based redaction; regex handles phones and common “Patient: …” patterns otherwise.

Medical dictionary

The sample dictionary in backend/data/medical_dictionary_sample.json includes common lab terms (e.g. Hemoglobin, Creatinine, TSH). You can extend it or replace it with NIH/UMLS/MeSH data; the ingest script expects JSON with term, definition, source, and optional category.

License

MIT.