🏥 Hospital RAG Assistant

An AI-powered Patient Query Assistant built using Retrieval-Augmented Generation (RAG). The system answers user queries strictly based on a provided hospital document.

---

🚀 Features

• 📄 PDF Upload & Ingestion
• ✂️ Smart Text Chunking
• 🧠 Semantic Search using Embeddings
• 🗄️ Vector Database (Supabase pgvector)
• 🤖 LLM-powered Answers (Groq - Llama 3.1)
• 🔍 Source Attribution (Page Numbers)
• 💬 Chat History (Conversational AI)
• 🖥️ Streamlit UI for interaction

---

🧠 Architecture

User (UI / API)
      ↓
FastAPI Backend
      ↓
PDF Upload → Text Extraction → Chunking
      ↓
Embeddings (SentenceTransformers)
      ↓
Supabase (pgvector)
      ↓
Query → Embedding → Similarity Search
      ↓
Top-K Chunks
      ↓
LLM (Groq - Llama 3.1)
      ↓
Final Answer + Sources

---

🛠️ Tech Stack

• FastAPI
• Streamlit
• SentenceTransformers (all-MiniLM-L6-v2)
• Supabase (PostgreSQL + pgvector)
• Groq API (Llama 3.1)

---

⚙️ Setup Instructions

1️⃣ Clone Repository

git clone https://github.com/your-username/hospital-rag-assistant.git
cd hospital-rag-assistant

---

2️⃣ Install Dependencies

pip install -r requirements.txt

Or manually:

pip install fastapi uvicorn streamlit sentence-transformers supabase groq python-dotenv python-multipart requests

---

3️⃣ Setup Environment Variables

Create .env file:

SUPABASE_URL=your_url
SUPABASE_KEY=your_secret_key
GROQ_API_KEY=your_groq_key

---

4️⃣ Run Backend

uvicorn app.main:app --reload

---

5️⃣ Run UI

python -m streamlit run app/ui.py

---

📌 API Endpoints

📄 Upload Document

POST /upload

💬 Ask Question

POST /query

Request:

{
  "question": "What is ICU cost?"
}

Response:

{
  "answer": "ICU cost is $600 per day.",
  "sources": ["page 5"]
}

---

🧪 Sample Queries

• What are OPD timings?
• Who is the cardiologist?
• What is the cost of MRI?
• What is ICU cost per day?
• What is the emergency number?

---

🚫 Important Constraint

The system answers ONLY from the document. If information is not found:

"I don't have that information in the provided document."

---

🎯 Key Highlights

• Prevents hallucination using RAG
• Supports conversational queries (chat history)
• Handles real-world document QA scenarios

---

👨‍💻 Author

Anshuman Singh