PDF Query Application 📚

A powerful RAG (Retrieval-Augmented Generation) application that allows you to upload PDF documents and ask questions about their content using AI. Built with Streamlit, LangChain, and Google Gemini.

Features ✨

• PDF Upload: Upload single or multiple PDF files
• Intelligent Question Answering: Ask questions about your PDF content and get accurate AI-generated answers
• Vector Search: Uses FAISS for efficient similarity search
• Local Embeddings: HuggingFace embeddings run locally (no API costs for embeddings)
• Google Gemini Integration: Powered by Google's Gemini 2.5 Flash model for answer generation
• User-Friendly Interface: Clean and intuitive Streamlit web interface

Technology Stack 🛠️

• Frontend: Streamlit
• LLM: Google Gemini 2.5 Flash (via LangChain)
• Embeddings: HuggingFace sentence-transformers/all-MiniLM-L6-v2
• Vector Store: FAISS
• PDF Processing: PyPDF
• Text Splitting: LangChain RecursiveCharacterTextSplitter

Prerequisites 📋

• Python 3.8 or higher
• Google API Key (for Gemini)

Installation 🚀

1. Clone the repository

   git clone <repository-url>
   cd pdfQuery

2. Create a virtual environment

   python -m venv venv
   source venv/bin/activate  # On Windows: venv\Scripts\activate

3. Install dependencies

   pip install -r requirements.txt

4. Set up environment variables

   Create a .env file in the project root:

   GOOGLE_API_KEY=your_google_api_key_here

   To get a Google API key:

   • Visit Google AI Studio
   • Create a new API key
   • Copy and paste it into your .env file

Usage 💡

1. Start the application

   streamlit run main.py

2. Upload PDFs

   • Click on the sidebar menu
   • Upload one or more PDF files
   • Click "Submit & Process" to process the documents

3. Ask Questions

   • Type your question in the text input field
   • The AI will search through your PDFs and provide relevant answers

How It Works 🔍

1. PDF Processing: Extracts text from uploaded PDF files
2. Text Chunking: Splits text into manageable chunks (1000 characters with 200 character overlap)
3. Embedding Generation: Creates vector embeddings using HuggingFace models (runs locally)
4. Vector Storage: Stores embeddings in FAISS index for fast similarity search
5. Question Answering:
   • Converts user question to embeddings
   • Searches for similar chunks in FAISS index
   • Sends relevant chunks to Google Gemini for answer generation
   • Returns AI-generated answer based on document context

Project Structure 📁

pdfQuery/
├── main.py              # Main application file
├── requirements.txt     # Python dependencies
├── .env                 # Environment variables (not in git)
├── .gitignore          # Git ignore file
├── README.md           # This file
├── faiss_index/        # FAISS vector database (generated)
└── venv/               # Virtual environment (not in git)

Configuration ⚙️

You can modify the following parameters in main.py:

• Chunk Size: Default is 1000 characters (line 31)
• Chunk Overlap: Default is 200 characters (line 32)
• Embedding Model: Default is sentence-transformers/all-MiniLM-L6-v2 (line 42)
• LLM Model: Default is gemini-2.5-flash (line 56)
• Temperature: Default is 0.3 for more focused answers (line 57)

Troubleshooting 🔧

Common Issues

1. "GOOGLE_API_KEY not found" error

   • Ensure your .env file exists and contains the API key
   • Verify the key is valid

2. "Please process the PDF first" error

   • Upload a PDF file first
   • Click "Submit & Process" before asking questions

3. Slow processing

   • First run downloads the HuggingFace model (~80MB)
   • Large PDFs take longer to process
   • Consider reducing chunk size for faster processing

4. Memory issues

   • For very large PDFs, consider processing them in batches
   • Reduce chunk size or increase chunk overlap

Dependencies 📦

Key dependencies include:

• streamlit - Web interface
• langchain - LLM framework
• langchain-google-genai - Google Gemini integration
• langchain-huggingface - HuggingFace embeddings
• faiss-cpu - Vector similarity search
• pypdf - PDF text extraction
• python-dotenv - Environment variable management

See requirements.txt for complete list.

Security 🔒

• Never commit your .env file or API keys to version control
• The .gitignore file is configured to exclude sensitive files
• Keep your Google API key secure and don't share it

Performance Tips 🚀

• Embeddings: Run locally on CPU, no API costs
• LLM Calls: Only made when asking questions (uses Google API)
• FAISS Index: Saved locally for faster subsequent queries
• Model Caching: HuggingFace models are cached after first download

Future Enhancements 🌟

Potential improvements:

• Support for more document formats (DOCX, TXT, etc.)
• Conversation history and context
• Multiple language support
• Custom embedding models
• Export answers to file
• Advanced search filters

---

Built with ❤️ using Streamlit, LangChain, and Google Gemini