PDF Chat Application

A Streamlit-based application that allows users to upload multiple PDF documents and have interactive conversations with their content using Google's Gemini AI.

Features

• 📚 Multiple PDF Support: Upload and process multiple PDF files simultaneously
• 💬 Interactive Chat: Ask questions about your documents and get intelligent responses
• 🧠 Memory: Maintains conversation history for context-aware responses
• ⚡ Fast Processing: Efficient text chunking and vector storage for quick retrieval
• 🎨 Clean UI: User-friendly Streamlit interface with custom styling

Prerequisites

• Python 3.7 or higher
• Google API key for Gemini AI

Installation

1. Clone or download the repository

2. Install required packages:

   pip install langchain langchain_community streamlit langchain_google_genai PyPDF2 python-dotenv chromadb

Usage

1. Start the application:

   streamlit run app.py

2. Upload PDFs:

   • Use the sidebar to upload one or more PDF files
   • Click the "Process" button to extract and process the text

3. Ask Questions:

   • Once processing is complete, use the text input to ask questions about your documents
   • The AI will provide responses based on the content of your PDFs

RAG Architecture

The application implements a Retrieval-Augmented Generation (RAG) pipeline:

graph LR
    %% Input
    PDF[📄 PDF Documents] --> EXTRACT[📖 Text Extraction<br/>PyPDF2]
    
    %% Document Processing
    EXTRACT --> CHUNK[✂️ Text Chunking<br/>CharacterTextSplitter]
    CHUNK --> EMBED[🧠 Embeddings<br/>Gemini Embedding]
    EMBED --> VECTOR[(🗄️ Vector Store<br/>ChromaDB)]
    
    %% Query Processing
    QUERY[❓ User Question] --> RETRIEVE[🔍 Similarity Search<br/>Vector Retrieval]
    VECTOR --> RETRIEVE
    
    %% Generation
    RETRIEVE --> CONTEXT[📋 Retrieved Context]
    CONTEXT --> LLM[🤖 Language Model<br/>Gemini 2.5 Flash]
    QUERY --> LLM
    MEMORY[🧩 Chat Memory] --> LLM
    
    %% Output
    LLM --> RESPONSE[💬 Generated Response]
    LLM --> MEMORY
    
    %% Styling
    classDef input fill:#e3f2fd,stroke:#1976d2,stroke-width:2px
    classDef process fill:#e8f5e8,stroke:#388e3c,stroke-width:2px
    classDef storage fill:#fff3e0,stroke:#f57c00,stroke-width:2px
    classDef generation fill:#fce4ec,stroke:#c2185b,stroke-width:2px
    classDef output fill:#f3e5f5,stroke:#7b1fa2,stroke-width:2px

    class PDF,QUERY input
    class EXTRACT,CHUNK,EMBED,RETRIEVE process
    class VECTOR,MEMORY storage
    class CONTEXT,LLM generation
    class RESPONSE output

Loading

How It Works

1. Text Extraction

• Uses PyPDF2 to extract text from uploaded PDF files
• Combines text from all pages across all uploaded documents

2. Text Chunking

• Splits the extracted text into manageable chunks (1000 characters each)
• Maintains 200-character overlap between chunks for context preservation

3. Vector Storage

• Creates embeddings using Google's Gemini embedding model
• Stores embeddings in a Chroma vector database for efficient similarity search

4. Conversational Chain

• Uses Google's Gemini 2.5 Flash model for generating responses
• Implements conversation memory to maintain context across questions
• Retrieves relevant document chunks to provide accurate, context-aware answers

Configuration

Model Settings

• LLM Model: gemini-2.5-flash (configurable in get_conversation_chain())
• Embedding Model: gemini-embedding-001 (configurable in get_vectorstore())
• Temperature: 0 (for more deterministic responses)

Text Processing Settings

• Chunk Size: 1000 characters
• Chunk Overlap: 200 characters
• Text Splitter: Character-based splitting on newlines

File Structure

project/
├── main.py                 # Main application file
├── html_templates.py      # HTML/CSS templates for chat UI
└── README.md             # This file

Dependencies

• streamlit: Web app framework
• langchain: LLM application framework
• langchain_community: Community extensions for LangChain
• langchain_google_genai: Google Gemini AI integration
• PyPDF2: PDF text extraction
• python-dotenv: Environment variable management
• chromadb: Vector database for embeddings

Troubleshooting

Common Issues

1. Google API Key Error:

   • Verify the API key has access to Gemini AI services

2. PDF Processing Issues:

   • Some PDFs may not extract text properly if they contain only images
   • Try using PDFs with selectable text

3. Memory Issues:

   • For very large documents, consider reducing chunk size or processing fewer files at once

4. Import Errors:

   • Ensure all required packages are installed with the correct versions
   • Try reinstalling packages if you encounter compatibility issues

Performance Tips

• Process PDFs one at a time for very large documents
• Clear browser cache if the app becomes unresponsive
• Restart the Streamlit server if you encounter persistent issues

Contributing

Feel free to fork this project and submit pull requests for improvements. Some areas for enhancement:

• Support for other document formats (Word, TXT, etc.)
• Advanced chunking strategies
• Multiple vector store options
• Improved error handling
• Additional LLM model options

License

This project is open source and available under the MIT License.