RAG Document Chat System

A modern web application that allows you to upload PDF documents and chat with them using AI. Built with Node.js, Express, and a beautiful responsive frontend.

Clean, modern interface ready for document upload

Features

• 📁 File Upload: Upload PDF documents with drag & drop support
• 🤖 AI Chat: Chat with your documents using OpenAI's GPT models
• 🗂️ Multiple Documents: Manage multiple documents with separate conversations
• 🎯 Namespace Isolation: Each document gets its own Pinecone namespace for accurate retrieval
• 💬 Conversation History: Maintain separate chat histories for each document
• 📱 Responsive Design: Beautiful, modern UI that works on all devices
• ⚡ Real-time Processing: Fast document processing and embedding generation
• 🌍 Multi-language Support: Ask questions in English, Vietnamese, or other languages
• 🔄 Auto-clean Data: Automatically clears old data on startup for fresh sessions

Tech Stack

Backend

• Node.js with Express.js
• LangChain for AI integration
• OpenAI for embeddings and chat
• Pinecone for vector database
• Multer for file uploads
• PDF-parse for document processing

Frontend

• Vanilla JavaScript with modern ES6+
• Responsive CSS with gradient design
• Drag & Drop file upload
• Real-time chat interface

Setup Instructions

1. Install Dependencies

npm install

2. Environment Variables

Create a .env file in the root directory with the following variables:

OPENAI_API_KEY=your_openai_api_key_here
PINECONE_API_KEY=your_pinecone_api_key_here
PORT=3001

3. Run the Application

# Start the server
npm start

# Or for development with auto-restart
npm run dev

The application will be available at http://localhost:3001

Demo

Example of chatting with a Vietnamese AWS Machine Learning document

How It Works

1. Document Upload: Users upload PDF files through the web interface
2. Processing: The system extracts text, chunks it, and generates embeddings
3. Storage: Embeddings are stored in Pinecone with unique namespaces per document
4. Chat: Users can select documents and ask questions
5. Retrieval: The system retrieves relevant chunks from the specific document's namespace
6. Generation: AI generates answers based on the retrieved context

API Endpoints

• GET /api/files - Get list of uploaded files
• POST /api/upload - Upload and process a PDF file
• POST /api/chat - Send a chat message for a specific document
• DELETE /api/files/:fileId - Delete a file and its data

File Structure

├── server.js              # Express server with API endpoints
├── public/
│   └── index.html         # Frontend application
├── src/
│   ├── chunk-texts.js     # Text chunking logic
│   ├── embed-texts.js     # Embedding generation
│   ├── generate-answer.js # AI answer generation
│   ├── parse-pdf.js       # PDF text extraction
│   └── vector-db.js       # Pinecone operations
├── uploads/               # Uploaded files storage
└── package.json

Usage

1. Upload Documents: Click "Upload Document" or drag & drop PDF files
2. Select Document: Click on any document in the sidebar to start chatting
3. Ask Questions: Type your questions in the chat input (supports multiple languages)
4. Switch Documents: Click different documents to have separate conversations
5. Delete Documents: Use the delete button to remove documents

Multi-language Support

The system supports cross-language queries:

• Upload a Vietnamese document and ask questions in English
• Upload an English document and ask questions in Vietnamese
• The AI will respond in the same language as your question

Configuration

The system uses the following default configurations:

• Chunk Size: 1000 characters
• Chunk Overlap: 200 characters
• Embedding Model: text-embedding-3-large (3072 dimensions)
• Chat Model: gpt-4o-mini
• Top K Results: 10 relevant chunks (with similarity filtering)
• File Size Limit: 10MB
• Similarity Threshold: 0.7 for cross-language search

Troubleshooting

Common Issues

1. "No matching chunks found": The document might not contain relevant information for your query
2. Upload fails: Check file size (max 10MB) and ensure it's a valid PDF
3. API errors: Verify your OpenAI and Pinecone API keys are correct
4. Dimension mismatch errors: The system automatically handles this by recreating the Pinecone index
5. Cross-language queries not working: Try asking in the same language as the document first

Environment Setup

Make sure you have:

• Node.js 16+ installed
• Valid OpenAI API key with sufficient credits
• Valid Pinecone API key with an active project

Original Terminal Version

The original terminal-based version is still available in index.js. You can run it with:

node index.js

This will:

• Create index if not exists
• Process sample PDF for the first time
• Accept queries from terminal and generate answers from PDF

License

MIT License - feel free to use this project for your own applications!