🧠 RAG-HEALTH: WHO-focused Retrieval-Augmented Generation System

project link: https://drive.google.com/file/d/1e15xKkPw1KtUhfN2a43LJ2b7T8nXV2RV/view?usp=drive_link

RAG-HEALTH is an end-to-end intelligent question-answering system designed for the healthcare domain. It leverages Retrieval-Augmented Generation (RAG) to retrieve information from trusted documents and generate accurate, explainable answers using state-of-the-art LLMs via the Groq API.

This system supports multimodal content (text, tables, and images) and is built to enable public access to healthcare knowledge using AI in a safe, transparent, and modular way.

---

🌐 Key Features

• 🔍 Vector-based document retrieval using ChromaDB and MiniLM embeddings
• 🤖 LLM-powered answer generation with Groq's LLaMA3 models
• 🖼️ Image-to-text captioning using ImgBB and Vision LLM
• 📊 PDF extraction for tables, text, and images
• ⚙️ Microservice architecture using FastAPI
• 🧩 Workflow orchestration using n8n

---

📁 Directory Structure

RAG-HEALTH/
│
├── data/                         # Source PDF files
├── helpers/                      # Core modules for embedding, retrieval, and LLM
│   ├── context_builder.py
│   ├── db_init.py
│   ├── image_llm.py
│   ├── prompt_builder.py
│   ├── text_llm.py
│
├── scripts/                      # Utility and automation scripts
│   ├── data_ingestion.py        # Extracts content and indexes it
│   ├── RAG_HEALTH.json          # n8n workflow config
│
├── services/                     # API layer (FastAPI microservices)
│   ├── retriever.py             # Context fetcher
│   ├── main.py                  # LLM responder
│
├── ui/                           # Streamlit frontend
│   ├── chat.py
│
├── vector_store/                # Persisted Chroma vector DB
├── .env                         # API keys and environment variables
├── requirements.txt
└── README.md

---

🧪 How It Works

1. PDF Ingestion:

   • scripts/data_ingestion.py extracts text, images, and tables.
   • Images are uploaded to ImgBB and captioned using Groq's LLM Vision model.
   • All content is split into semantic chunks and stored in Chroma with metadata.

2. Query Flow:

   • User enters a healthcare query (e.g., "What are sterilization guidelines before surgery?").
   • The retriever service searches Chroma for relevant document chunks.
   • The main service sends the prompt to Groq LLaMA3 API and returns the answer.

3. Multimodal Understanding:

   • Tables and images from PDFs are treated as context via captions or markdown rendering.

4. n8n Workflow:

   • The entire query → retrieval → response flow is orchestrated using n8n.

---

🔁 n8n Workflow (Visual Overview)

Trigger Node (Webhook /query)
    ↓
HTTP Node → POST /retrieve
    ↓
HTTP Node → POST /ask
    ↓
Respond to Webhook

• Webhook Endpoint: /query
• Expected Body:

{
  "query": "Explain laparoscopic sterilization steps."
}

• Intermediate Responses:
  • /retrieve → { context, question }
  • /ask → { answer, sources, context }

---

💻 Running Locally

1. Clone the Repository

git clone https://github.com/yourname/RAG-HEALTH
cd RAG-HEALTH

2. Install Python Dependencies

pip install -r requirements.txt

3. Configure Environment

Create a .env file with the following:

GROQ_API_KEY=your_groq_key
IMGBB_API_KEY=your_imgbb_key
HUGGINGFACEHUB_API_TOKEN=your_huggingface_token
CHROMA_DB_DIR=./vector_store

4. Ingest Documents

python scripts/data_ingestion.py

5. Start Backend Services

# In one terminal
uvicorn services.retriever:app --port 8000

# In another terminal
uvicorn services.main:app --port 8001

6. Start n8n

• Import scripts/RAG_HEALTH.json into your n8n dashboard
• Start the workflow manually or via API trigger

7. (Optional) Launch Streamlit UI

streamlit run ui/chat.py

---

🤖 Technologies Used

Layer	Technology
LLM	Groq LLaMA3-70B via OpenAI API
Embeddings	HuggingFace all-MiniLM-L6-v2
Vector Store	Chroma DB
Backend	FastAPI
Frontend	Streamlit
Image Upload	ImgBB
Orchestration	n8n
Parsing PDFs	PyMuPDF (fitz), pdfplumber

---

🧠 Prompt Design

[Context]
- Chunked and cleaned healthcare-relevant paragraphs
- Captions from PDF images and tables

[Question]
12 Healthy habits?.

[Answer]
(Generated using Groq LLM)

---

🔮 Future Enhancements

• Image + table support
• Text-to-prompt automation
• Session-level memory
• Incremental PDF updates
• Integrated query analytics dashboard
• Slack / Discord bot interface
• Speech-to-text input with Whisper

---

🧑‍⚕️ Ideal Use Cases

• WHO guidelines
• Precautions during catastrophe

---

📃 License

This project is licensed under the MIT License.

---