🦙 Local Llama Chat with Document Intelligence

GenAI companion for document processing, Q&A. fun for first time genai developers/researchers, students, and professionals who want to chat with their documents.

🎨 Features at a Glance

Feature	Description
📄 Document Q&A	Chat with PDFs, Word, PowerPoint, images
🌐 Web Integration	Scrape websites and YouTube videos
🦙 Local LLM	Ollama-powered conversations
🔧 Smart Tools	Search, math, weather, analytics
📊 Analytics	Track response quality and improve
🗃️ Database	PostgreSQL integration for structured data

1. document loaders in document_processor.py
2. Create new tools in tools.py
3. ** dependencies** in requirements.txt
4. Add tests in the tests/ directory
5. Improve documentation in this README

📄 License & Acknowledgments

This project is designed for educational and development purposes, showcasing the use of local AI combined with basic document intelligence.

Document Support

• 📄 Documents: PDF, TXT, CSV, JSON, DOCX, XLSX, Markdown
• 🎨 Presentations: PowerPoint (PPTX) with full content extraction
• 🖼️ Images: JPG, PNG, GIF, BMP with OCR capabilities
• 🌐 Web Content: Scrape and analyze web pages automatically
• 🎥 YouTube: Extract and search through video transcripts
• �️ Databases: Direct PostgreSQL integration for structured data

🏠 Local

Everything runs on your machine using Ollama. No data leaves your system, no API keys required, no subscription fees. Your documents, your conversations, your privacy.

🚀 Getting Started

Quick Setup

# Clone and dive in
git clone <This Repository>
cd <Your Project Directory>
python -m venv venv && source venv/bin/activate
pip install -r requirements.txt

# Start Ollama (if not running)
ollama serve
ollama pull llama3.2:1b

# Launch your AI companion
streamlit run app.py --server.port=8501 # (setting the port is optional)

Environment Setup

Copy .env.example to .env and configure:

OLLAMA_HOST=http://localhost:11434
OLLAMA_MODEL=llama3.2:1b
STREAMLIT_SERVER_PORT=8501
DATABASE_URL=postgresql://user@localhost:5432/dbname  # Optional

🎯 How to Use

1. Upload Files - Drag & drop PDFs, Word docs, slides, images
2. Process & Analyze - Click "📁 Process Documents"
3. Ask Anything - "What are the key themes across all documents?"
4. Get Smart Answers - Cross-document insights with source attribution

🌐 Web & YouTube Integration

1. Web Pages - Paste public URL and extract content
2. YouTube Videos - Add video links to search through transcripts
3. Mixed Sources - Combine documents, web, and video content
4. Unified Search - Ask questions across all content types

🧪 Testing & Debugging

Explore the test suite to check functionality:

# Document processing demo
streamlit run tests/test_document_processing.py --server.port=8502

# Upload debugging
streamlit run tests/debug_upload.py --server.port=8503

# Similarity search testing
streamlit run tests/test_similarity.py --server.port=8504

# Feedback system demo
streamlit run tests/test_feedback.py --server.port=8505

📁 Project Architecture

windsurf-project/
├── app.py                    # 🎯 Main application
├── document_processor.py     # 📚 Document intelligence core
├── langchain_integration.py  # 🦙 LLM management
├── tools.py                  # 🔧 Utility tools
├── feedback_analytics.py     # 📊 Quality tracking
├── requirements.txt          # 📦 Dependencies
├── .env.example             # 🔧 Configuration template
├── tests/                   # 🧪 Test suite
└── README.md                # 📖 This guide

⚡ Tips

🏎️ Speed Optimisation

• Use llama3.2:1b for lightning-fast responses
• Enable Metal GPU running the shell script will start the application with metal=1
• Limit conversations to 5-10 messages for best performance
• Process documents in batches for large files

💾 Memory Management

• Clear processed documents when switching projects
• Use smaller chunks for very large documents
• Restart periodically for long sessions

🐛 Friendly Troubleshooting

🦙 Ollama Issues

# Check if Ollama is running
ollama list

# Start the server
ollama serve

# Pull recommended model
ollama pull llama3.2:1b

📄 Document Processing

• Supported formats: PDF, TXT, CSV, JSON, DOCX, XLSX, MD, PPTX, JPG, PNG
• File size limit: 200MB per file (configurable via the .streamlit/config.toml file)
• Multiple files: Upload and process simultaneously
• Cross-document Q&A: Ask questions spanning all uploaded content

🌐 Network Issues

• Web scraping: Works with most public websites
• YouTube transcripts: Automatic extraction for videos with captions
• No API keys: Everything works with free, public APIs

Special thanks to:

• Streamlit for the web framework
• Ollama for making local LLMs accessible
• LangChain for powerful document processing
• DuckDuckGo for privacy-focused web search