Document Assistant RAG

A sophisticated Retrieval Augmented Generation (RAG) system that allows users to upload documents and ask questions about their content with intelligent scope validation.

Features

🚀 Core Capabilities

• Multi-format Document Support: PDF, DOCX, TXT, and Markdown files
• Advanced Text Chunking: Semantic, recursive character, and sentence-based splitting
• High-Quality Embeddings: Using Sentence Transformers (all-MiniLM-L6-v2)
• Vector Database: ChromaDB for efficient similarity search
• Multiple LLM Options: OpenAI GPT and Hugging Face models
• Scope Validation: Intelligent detection of out-of-scope queries
• Interactive Web Interface: Streamlit-based user-friendly interface

🧠 Advanced Features

• Content-Aware Chunking: Different strategies for code, tables, and paragraphs
• Context Preservation: Overlapping chunks maintain context continuity
• Semantic Search: Find relevant information using meaning, not just keywords
• Source Attribution: See which documents and sections answers came from
• Error Handling: Robust processing with graceful fallbacks

Installation

Prerequisites

• Python 3.8 or higher
• pip package manager

Setup Instructions

1. Clone the repository:

   git clone <repository-url>
   cd CodeLLM

2. Create a virtual environment (recommended):

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

3. Install dependencies:

   pip install -r requirements.txt

4. Optional: Set up OpenAI API (for GPT models):

   export OPENAI_API_KEY="your-api-key-here"
   # Or create a .env file with: OPENAI_API_KEY=your-api-key-here

Usage

Starting the Application

Run the Streamlit application:

streamlit run app.py

The application will open in your web browser at http://localhost:8501

Using the Document Assistant

1. Upload Documents:

   • Use the sidebar to upload PDF, DOCX, TXT, or Markdown files
   • Multiple files can be uploaded simultaneously
   • Processing happens automatically

2. Ask Questions:

   • Type questions about your uploaded documents in the chat interface
   • Get contextual answers with source attributions
   • Receive warnings for out-of-scope questions

3. Review Sources:

   • Click on source expandables to see which document sections were used
   • Understand the context behind each answer

Example Queries

Good (In-scope) Questions:

• "What are the main points discussed in the document?"
• "Can you summarize the key findings from Chapter 3?"
• "What does the author say about [specific topic]?"
• "According to the document, how does [process] work?"

Out-of-scope Questions (will be flagged):

• "What's the weather today?"
• "Can you recommend a good restaurant?"
• "What's the latest news about [unrelated topic]?"

Architecture

Core Components

1. RAG Engine (backend/rag_engine.py)

   • Orchestrates the entire RAG pipeline
   • Manages embeddings and vector storage
   • Integrates with multiple LLM providers

2. Document Processor (backend/document_processor.py)

   • Extracts text from various document formats
   • Handles encoding detection and text cleaning
   • Robust error handling for corrupted files

3. Text Chunker (backend/text_chunker.py)

   • Advanced chunking strategies based on content type
   • Preserves context with intelligent overlapping
   • Handles code, tables, and regular text differently

4. Scope Validator (backend/scope_validator.py)

   • Validates query relevance using multiple methods
   • Semantic similarity and keyword analysis
   • Pattern-based out-of-scope detection

Technology Stack

• Frontend: Streamlit for interactive web interface
• Embeddings: Sentence Transformers (all-MiniLM-L6-v2)
• Vector Database: ChromaDB with persistent storage
• LLM Options:
  • OpenAI GPT-3.5/4 (requires API key)
  • Hugging Face Transformers (free, local)
• Document Processing: PyPDF2, pdfplumber, python-docx
• Text Processing: Advanced regex and NLP techniques

Configuration

Environment Variables

Create a .env file in the project root:

# OpenAI API (optional, for GPT models)
OPENAI_API_KEY=your-openai-api-key

# Chunking Configuration
CHUNK_SIZE=1000
CHUNK_OVERLAP=200
MIN_CHUNK_SIZE=100

# Embedding Model
EMBEDDING_MODEL=all-MiniLM-L6-v2

# LLM Configuration
LLM_PROVIDER=huggingface  # or "openai"
LLM_MODEL=microsoft/DialoGPT-small

Customization Options

You can customize the RAG system by modifying parameters in the code:

• Chunk sizes: Adjust chunk_size and chunk_overlap in text_chunker.py
• Embedding model: Change the model in rag_engine.py
• LLM provider: Switch between OpenAI and Hugging Face models
• Similarity thresholds: Tune scope validation sensitivity

Performance Tips

For Better Results

1. Upload high-quality documents: Clear text, good formatting
2. Use specific questions: More specific queries get better answers
3. Check document scope: Ensure questions relate to uploaded content
4. Multiple documents: Upload related documents for comprehensive coverage

System Requirements

• Memory: 4GB+ RAM recommended (8GB+ for large documents)
• Storage: Vector database grows with document size
• CPU: Modern multi-core processor for faster processing
• GPU: Optional, improves Hugging Face model performance

Troubleshooting

Common Issues

1. Import Errors:

   pip install --upgrade -r requirements.txt

2. PDF Processing Fails:

   pip install pymupdf  # Alternative PDF processor

3. Memory Issues with Large Documents:

   • Reduce chunk size
   • Process documents individually
   • Use a machine with more RAM

4. Slow Performance:

   • Use GPU acceleration if available
   • Switch to lighter embedding models
   • Reduce the number of retrieved chunks

Logs and Debugging

The application logs important information to help with debugging:

• Document processing status
• Chunking statistics
• Query processing details
• Error messages with context

Contributing

We welcome contributions! Here's how to get started:

1. Fork the repository
2. Create a feature branch
3. Make your changes
4. Add tests if applicable
5. Submit a pull request

Development Setup

# Install development dependencies
pip install -r requirements.txt
pip install pytest black flake8

# Run tests
pytest

# Format code
black .

# Check code style
flake8 .

License

This project is licensed under the MIT License - see the LICENSE file for details.

Acknowledgments

• Sentence Transformers: For high-quality embeddings
• ChromaDB: For efficient vector storage
• Streamlit: For the interactive web interface
• Hugging Face: For free, accessible LLM models
• OpenAI: For advanced language understanding capabilities

Support

If you encounter issues or have questions:

1. Check the troubleshooting section above
2. Search existing GitHub issues
3. Create a new issue with detailed information
4. Include error logs and system information

---

Built with ❤️ for better document understanding and knowledge extraction# CODELLM