PDF-Based AI Chat System

A Node.js application that creates an intelligent chat system capable of answering questions about PDF documents using AI embeddings and vector search.

📺 Tutorial

This project was built following this YouTube tutorial: PDF-Based AI Chat System Tutorial

🚀 Features

• PDF Processing: Load and chunk PDF documents for better processing
• Vector Embeddings: Convert text chunks into vector embeddings using Google's Generative AI
• Vector Database: Store embeddings in Pinecone for efficient similarity search
• Intelligent Chat: Interactive chat system that answers questions based on PDF content
• Context-Aware: Maintains conversation history for better context understanding
• Query Transformation: Rewrites follow-up questions for better search results

📋 Prerequisites

• Node.js (v16 or higher)
• npm or yarn
• Google Gemini API key
• Pinecone account and API key

🛠️ Installation

1. Clone this repository:

git clone <your-repo-url>
cd Day08

2. Install dependencies:

npm install

3. Set up environment variables: Create a .env file in the root directory with:

GEMINI_API_KEY=your_gemini_api_key_here
PINECONE_API_KEY=your_pinecone_api_key_here
PINECONE_ENVIRONMENT=your_pinecone_environment
PINECONE_INDEX_NAME=your_pinecone_index_name

📂 Project Structure

├── index.js          # PDF processing and vector storage script
├── query.js           # Interactive chat interface
├── dsa.pdf           # Sample PDF document (Data Structures & Algorithms)
├── package.json      # Project dependencies
├── .env              # Environment variables (not tracked in git)
└── README.md         # Project documentation

🎯 Usage

1. Process PDF and Store Embeddings

First, run the indexing script to process your PDF and store it in Pinecone:

node index.js

This will:

• Load the PDF document (dsa.pdf)
• Split it into manageable chunks
• Generate vector embeddings
• Store everything in Pinecone database

2. Start Interactive Chat

Once indexing is complete, start the chat interface:

node query.js

The system will prompt you to ask questions about the PDF content.

🔧 Dependencies

Core Dependencies

• @google/genai: Google Generative AI client
• @langchain/community: LangChain community modules
• @langchain/google-genai: Google AI integration for LangChain
• @langchain/pinecone: Pinecone integration for LangChain
• @langchain/textsplitters: Text splitting utilities
• @pinecone-database/pinecone: Pinecone database client
• dotenv: Environment variable management
• readline-sync: Synchronous user input

🏗️ How It Works

1. Document Processing (index.js):

   • Loads PDF using PDFLoader
   • Splits text into chunks with overlap for better context
   • Converts chunks to vector embeddings using Google's text-embedding-004 model
   • Stores embeddings in Pinecone with metadata

2. Query Processing (query.js):

   • Takes user input and transforms follow-up questions for clarity
   • Converts questions to vector embeddings
   • Performs similarity search in Pinecone to find relevant content
   • Uses retrieved context to generate answers via Gemini AI
   • Maintains conversation history for context-aware responses

🔑 Key Features

• Chunking Strategy: Uses RecursiveCharacterTextSplitter with 1000 character chunks and 200 character overlap
• Embedding Model: Google's text-embedding-004 for high-quality vector representations
• Vector Search: Top-K similarity search (K=10) for relevant context retrieval
• AI Model: Gemini-2.0-flash for generating human-like responses
• Error Handling: Graceful handling when answers aren't found in the document

🚨 Important Notes

• Ensure your PDF file is placed in the root directory as dsa.pdf or update the path in index.js
• Run index.js before query.js to populate the vector database
• The system is configured for Data Structure and Algorithm content but can be adapted for other domains
• Keep your API keys secure and never commit them to version control

🤝 Contributing

1. Fork the repository
2. Create a feature branch
3. Make your changes
4. Test thoroughly
5. Submit a pull request

� Tutorial

This project was built following this YouTube tutorial: PDF-Based AI Chat System Tutorial

🐛 Troubleshooting

Common Issues

1. Module Type Error: Ensure "type": "module" is set in package.json
2. API Key Issues: Verify your environment variables are correctly set
3. Dependency Conflicts: Use npm install --legacy-peer-deps if needed
4. Pinecone Connection: Ensure your Pinecone index exists and is accessible

Support

If you encounter any issues, please check:

• Node.js version compatibility
• API key validity
• Network connectivity
• Pinecone index configuration