🏥 MediChat Pro

An AI-powered Medical Document Assistant that processes patient history PDFs and enables intelligent, RAG-based conversational querying — built with a production-ready FastAPI + Streamlit architecture.

🌐 Live Demo

Check the deployed application here:

Open MediChat Pro

---

📌 Project Overview

MediChat Pro allows users to upload patient medical history PDFs and chat with the data in natural language. The system retrieves relevant chunks from an embedded vector store and generates structured, safety-aware responses using an LLM.

Built as a functional RAG system.

---

🏗️ Architecture

┌──────────────────────────────────────────┐
│              User (Browser)              │
└──────────────────┬───────────────────────┘
                   │
       ┌───────────▼───────────┐
       │  Streamlit Frontend   │  main.py  :8501
       └───────────┬───────────┘
                   │ HTTP REST
       ┌───────────▼───────────┐
       │   FastAPI Backend     │  api.py    :8000
       └──┬──────────┬─────────┘
          │          │          │
   ┌──────▼──┐  ┌────▼──┐  ┌───▼──────┐
   │ChromaDB │  │  S3   │  │          │
   │(Vectors)│  │(PDFs) │  │  (LLM)   │
   └─────────┘  └───────┘  └──────────┘

Key design decision: FastAPI and Streamlit run as separate services for independent deployment, testability, and scalability.

---

🚀 Getting Started

Prerequisites

• Python 3.10+
• ChromaDB Cloud account
• Euri AI API key / OpenAI API key
• HuggingFace API token
• AWS S3 bucket (optional)

Installation

git clone <repo-url>
cd medichat-pro
pip install -r requirements.txt

Environment Variables

Create a .env file:

EURI_API_KEY=your_euri_api_key
LLM_MODEL=gpt-4.1-nano

HF_API_TOKEN=your_huggingface_token
EMBEDDING_MODEL=sentence-transformers/all-MiniLM-L6-v2

CHROMA_API_KEY=your_chroma_api_key
CHROMA_TENANT=your_tenant
CHROMA_DATABASE=medibot
CHROMA_COLLECTION=medical_documents

S3_ACCESS_KEY=your_aws_access_key
S3_SECRET_KEY=your_aws_secret_key
S3_BUCKET_NAME=medibot-euron-2025
S3_REGION=ap-south-1

EMAIL_SMTP_SERVER=smtp.gmail.com
EMAIL_SMTP_PORT=587
EMAIL_SENDER=your_email@gmail.com
EMAIL_PASSWORD=your_app_password
EMAIL_RECEIVER=default_receiver@example.com

Running

# Terminal 1 — Backend
uvicorn api:app --reload --host 0.0.0.0 --port 8000

# Terminal 2 — Frontend
streamlit run main.py --server.address 0.0.0.0 --server.port 8501

Open http://localhost:8501

---

📂 Project Structure

medichat-pro/
├── api.py                    # FastAPI backend — 12 REST endpoints
├── main.py                  # Streamlit frontend — chat UI
├── app/
│   ├── config.py             # Env variable loading
│   ├── pdf_utils.py          # PDF extraction + cleaning
│   ├── vectorstore_utils.py  # ChromaDB + HuggingFace embeddings
│   ├── chat_utils.py         # Euri AI LLM client
│   ├── email_utils.py        # SMTP email reports
│   └── s3_utils.py           # AWS S3 management
├── requirements.txt
└── README.md

---

🔌 API Reference

Method	Endpoint	Description
GET	/health	Health check
GET	/status	Vectorstore state
POST	/init	Load from ChromaDB
POST	/documents/upload	Upload PDFs → embed
POST	/documents/process-s3	Re-embed from S3
GET	/documents/s3-list	List S3 PDFs
DELETE	/documents/clear	Wipe ChromaDB
POST	/chat	RAG chat + insights
POST	/email/send-report	Send analytics email
POST	/email/support-ticket	Create support ticket
POST	/email/save-session	Save session email
POST	/email/test	SMTP test

---

💬 Chat Commands

Type in chat	Action
send report to user@email.com	Email full analysis
support ticket	Create support ticket
process s3	Embed all S3 docs
save session	Email session summary

---

📊 RAG Insight Engine

Every response includes a RAG Quality Panel with 9 patient-safety metrics:

Metric	What it measures
Confidence Score	Weighted composite of all below (0–100%)
Patient Match Score	% chunks containing the queried patient's name
Field Coverage	Standard medical fields found in retrieved context
Entity Faithfulness	Medical entities from context echoed in response
Critical Field Safety	Allergy/medication/diagnosis always surfaced
Contamination Risk	Multiple patients' data mixing in chunks
Age/Name Consistency	Age in chunks vs. age stated in response
Intent Coverage	Response addresses what was actually asked
Avg Chunk Relevance	Token-overlap between query and each chunk

---

🛡️ Patient Safety Design

1. Allergy First — Allergy info always surfaced; if missing, explicitly flagged
2. Patient Isolation — Contamination metric detects cross-patient data leakage
3. No Hallucination — LLM prompted to never invent medication details
4. Missing Data Protocol — Absent fields stated explicitly, never implied normal

---

⚙️ Tech Stack

Layer	Technology
Frontend	Streamlit
Backend	FastAPI
Vector Store	ChromaDB Cloud 1.0
Embeddings	HuggingFace all-MiniLM-L6-v2
LLM	gpt-4.1-nano
Text Splitting	LangChain RecursiveCharacterTextSplitter
PDF Parsing	pypdf 3.17
Storage	AWS S3
Email	Python SMTP (Gmail)

---

🚢 Deployment (Render)

# Backend start command
uvicorn api:app --host 0.0.0.0 --port $PORT

# Frontend start command
streamlit run main2.py --server.address 0.0.0.0 --server.port $PORT

Set API_BASE_URL environment variable on the Streamlit service pointing to your deployed FastAPI URL.

---

🔮 Roadmap

• RAGAS evaluation framework
• CrossEncoder re-ranking
• pytest test coverage
• Section-aware chunking
• Per-user document isolation (JWT + metadata filters)

---

⚠️ For informational and educational purposes only. Not a substitute for professional medical advice.