---

Medical RAG (Retrieval-Augmented Generation) Knowledge Base

This project builds a Medical Retrieval-Augmented Generation (RAG) Knowledge Base using Next.js, LangChain, and Pinecone as the vector database. The system processes medical documents (e.g., PDF files) and stores their embeddings in a vector database, enabling efficient similarity search to retrieve relevant information in response to user queries.

Table of Contents

• Overview
• Features
• Technologies
• Setup and Installation
• Configuration
• Usage
• Project Structure
• Troubleshooting

Overview

This application provides a user-friendly interface for uploading medical documents such as PDFs, converting them into chunks, and storing their embeddings in a vector database (Pinecone). The system uses LangChain for chunking documents and Hugging Face's embedding models to generate embeddings. These embeddings can be searched efficiently for retrieving relevant information from medical knowledge bases.

Features

• Upload and process medical documents (PDF, text files).
• Chunk large documents and generate vector embeddings using an ONNX-based model.
• Store document embeddings in Pinecone for efficient similarity search.
• Progress bar for tracking the embedding process.
• Easily extendable to include more medical documents and enhance the knowledge base.

Technologies

• Next.js: For building the front-end and API routes.
• LangChain: For document chunking and embedding.
• Pinecone: Vector database for storing document embeddings.
• Hugging Face Transformers: To generate embeddings using pre-trained models.
• TypeScript: For type safety.
• Shadcn UI: For building the UI components.

Setup and Installation

1. Clone the Repository

git clone https://github.com/ShantamShukla/medicalrag.git
cd medicalrag

2. Install Dependencies

Ensure you have Node.js (version 18 or above) installed. Then install the required packages:

npm install

3. Environment Variables

Create a .env file in the root of the project and add the following environment variables:

PINECONE_API_KEY=your_pinecone_api_key

You can obtain your Pinecone API key by signing up at Pinecone.

4. Set Up Pinecone

• Log in to Pinecone and create an index.
• Use 1024 as the dimension for embeddings, and cosine similarity as the metric.

5. Run the Application

After setting up, run the development server:

npm run dev

Your app will now be running at http://localhost:3000.

Configuration

The following configurations are essential for the app:

• Pinecone Index Name: The name of the index in which document embeddings are stored.
• Namespace: You can store different document sets in Pinecone under separate namespaces.

You can specify these in the UI before uploading files.

Usage

1. Upload Documents

• Open the app at http://localhost:3000.
• Select PDF or text files from your local system.
• Specify the index name and namespace for storing embeddings.
• Click the Upload button to begin the embedding process.

The progress bar will track the process, and embeddings will be pushed to the Pinecone database.

2. View Stored Vectors

Once the embeddings are processed and stored, you can view them in your Pinecone dashboard.

3. Searching Embeddings (Future Feature)

You can extend the app by building a search interface to query the stored embeddings and retrieve relevant information.

Project Structure

├── pages/
│   ├── api/
│   │   └── updatedatabase.ts  # API route to handle document uploading and embedding
│   ├── index.tsx              # Main page for uploading documents
├── utils/
│   ├── updateVectorDB.ts      # Utility function to chunk, embed, and upload vectors
├── components/
│   └── UI components for the project
├── documents/                 # Place for storing documents
├── config.ts                  # Configuration file (batch sizes, etc.)
├── .env                       # Environment variables
└── README.md                  # This readme file

Troubleshooting

1. Fetch Errors

If you encounter the error:

Failed to find any user-provided fetch implementation. Using global fetch implementation.

Ensure you are running Node.js version 18+, which provides a global fetch function. You can update Node.js using:

nvm install 18
nvm use 18

2. Missing API Key

If the vectors are not being pushed to Pinecone, ensure your API key is correctly configured in the .env file.

3. Large File Processing

For large PDF documents, the embedding process may take time. Let it run until the progress bar shows completion.

---