🏥 Digital Medical Representative AI

An intelligent agentic RAG system that provides healthcare professionals with instant, accurate drug and reimbursement information from verified Indian sources.

✨ Features

• Agentic RAG - Auto-classifies queries and routes to appropriate document collections
• Cited Answers - Every response includes [Source: Document Name, Page: X]
• Trust-First Design - Returns "Not found in verified sources" when unsure
• Multi-Upload - Batch upload up to 5 medical documents at once
• Role-Based Access - Admin-only document management

📋 Document Categories

Category	Content
APPROVAL	CDSCO drug labels, indications, dosage
SAFETY	Contraindications, side effects, warnings
REIMBURSEMENT	Ayushman Bharat coverage, PMJAY eligibility

---

🚀 Quick Start

Prerequisites

• Node.js 18-20
• Docker (for Qdrant)
• MongoDB
• HuggingFace API Token

1. Clone & Install

git clone <repo-url>
cd Udemy_project
git checkout feature/rag
npm install

2. Environment Setup

Create a .env file (see .env.example):

# Database
MONGO_URI=mongodb://localhost:27017/medical-rep

# Server
PORT=8000
CORS_ORIGIN=*

# JWT Auth
ACCESS_TOKEN_SECRET=your-secret-key
ACCESS_TOKEN_EXPIRY=10m
REFRESH_TOKEN_SECRET=your-refresh-secret
REFRESH_TOKEN_EXPIRY=1d

# RAG Configuration
HF_TOKEN=your-huggingface-token
QDRANT_URL=http://localhost:6333

3. Start Qdrant Vector Database

cd docker
docker compose up -d

4. Run the Server

npm run dev

---

👤 Setting Up Admin User

After registering a user, promote them to admin:

// In MongoDB shell
db.users.updateOne(
  { email: "admin@example.com" },
  { $set: { systemRole: "ADMIN" } }
)

---

📡 API Endpoints

Authentication

Method	Endpoint	Description
POST	/api/v1/auth/register	Register new user
POST	/api/v1/auth/login	Login

RAG System

Method	Endpoint	Auth	Description
GET	/api/v1/rag/health	Public	System health check
GET	/api/v1/rag/documents	User	List all documents
GET	/api/v1/rag/documents/:id	User	Get document details
POST	/api/v1/rag/chat	User	Chat with AI
POST	/api/v1/rag/upload	Admin	Upload single PDF
POST	/api/v1/rag/upload-multiple	Admin	Upload multiple PDFs
DELETE	/api/v1/rag/documents/:id	Admin	Delete document

---

📖 Usage Examples

Upload a Document (Admin)

curl -X POST http://localhost:8000/api/v1/rag/upload \
  -H "Authorization: Bearer YOUR_TOKEN" \
  -F "document=@drug_label.pdf" \
  -F "category=APPROVAL" \
  -F "source=CDSCO"

Upload Multiple Documents (Admin)

curl -X POST http://localhost:8000/api/v1/rag/upload-multiple \
  -H "Authorization: Bearer YOUR_TOKEN" \
  -F "documents=@drug1.pdf" \
  -F "documents=@drug2.pdf" \
  -F "category=SAFETY" \
  -F "source=CDSCO"

Chat with AI

curl -X POST http://localhost:8000/api/v1/rag/chat \
  -H "Authorization: Bearer YOUR_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"query": "Is paracetamol approved for fever in India?"}'

Example Response:

{
  "statusCode": 200,
  "data": {
    "answer": "Yes, paracetamol is approved for fever treatment in India. [Source: Paracetamol Label, Page: 1]",
    "sources": [...],
    "classification": {
      "categories": ["APPROVAL"],
      "primaryCategory": "APPROVAL",
      "confidence": "high"
    },
    "foundInSources": true
  }
}

Query Without Match

{
  "answer": "Information not found in verified Indian sources. Please consult official CDSCO or healthcare provider resources.",
  "foundInSources": false
}

---

🏗️ Project Structure

src/
├── controllers/
│   └── rag.controller.js       # Upload, chat, document management
├── middlewares/
│   ├── auth.middleware.js      # JWT verification
│   └── admin.middleware.js     # Admin role check
├── models/
│   ├── user.model.js           # User with systemRole
│   └── medicalDocument.model.js # Document metadata
├── routes/
│   └── rag.routes.js           # RAG API routes
├── services/
│   ├── vectorStore.service.js  # PDF indexing to Qdrant
│   ├── retriever.service.js    # Agentic RAG chat
│   └── queryClassifier.service.js # Query routing
└── utils/
    ├── constants.js            # Document categories, roles
    └── multer.config.js        # PDF upload config

---

� What This System Does / Doesn't Do

✅ Does

• Provide factual drug information from verified documents
• Show approval status, contraindications, dosage
• Check Ayushman Bharat reimbursement eligibility
• Cite all sources with page numbers

❌ Doesn't

• Provide medical advice or diagnosis
• Recommend treatments
• Make claims without source documents
• Answer questions not in uploaded documents

---

📝 Recommended Documents for Demo

1. CDSCO Drug Label - Any approved drug's official label (PDF)
2. Clinical Guideline - Indian medical association guidelines
3. Ayushman Bharat Document - PMJAY coverage/pricing info

---

🛠️ Tech Stack

• Backend: Express.js (ES Modules)
• Database: MongoDB + Mongoose
• Vector Store: Qdrant
• Embeddings: HuggingFace BAAI/bge-base-en-v1.5
• LLM: Qwen via HuggingFace Inference
• File Upload: Multer
• Auth: JWT

---

📄 License

ISC

---

👨‍💻 Author

Digvijay Mukund