📚 Intelligent Document Summarizer

A powerful Streamlit-based application that uses LangChain and Ollama to generate intelligent summaries from multiple document formats. Upload documents, provide custom instructions, and get professionally formatted summaries tailored to your needs.

✨ Features

• Multi-Format Support: Load and process PDF, DOCX, PPTX, TXT, and HTML documents
• Custom Instructions: Control summarization style through natural language (e.g., "Teach this to a beginner", "Extract key financial risks")
• Multiple Summarization Methods:
  • Stuff: Fast, for smaller documents
  • Map Reduce: Scalable, processes large documents in chunks
  • Refine: Iterative refinement for comprehensive summaries
• Chat Interface: Interactive conversation-style interface with chat history
• Real-time Processing: Live feedback during document loading and summarization
• Clean Markdown Output: Professionally formatted summaries with headers, bullets, and tables

🚀 Quick Start

Prerequisites

• Python 3.10 or higher
• Ollama installed and running locally
• A compatible Ollama model (default: mistral-large-3:675b-cloud)

Installation

1. Clone or navigate to the project directory:

   cd /path/to/summarization

2. Install dependencies using uv (recommended):

   uv sync

   Or with pip:

   pip install -r requirements.txt

3. Ensure Ollama is running:

   ollama serve

4. Pull your desired model (if not already available):

   ollama pull mistral-large-3:675b-cloud
   # Or use another model like:
   # ollama pull llama3
   # ollama pull mistral

Running the App

uv run streamlit run main.py

Or with standard Python:

streamlit run main.py

The app will open in your browser at http://localhost:8501.

📖 Usage Guide

Step 1: Configure Settings (Sidebar)

1. Ollama Model: Enter the model name (default: mistral-large-3:675b-cloud)
2. Method: Choose summarization approach:
   • Stuff - Best for documents under ~4000 tokens
   • Map Reduce - Best for large documents
   • Refine - Best for iterative, comprehensive summaries

Step 2: Upload Documents

1. Click "Upload Files" in the sidebar
2. Select one or more documents (PDF, DOCX, PPTX, TXT, or HTML)
3. Click "Process Documents"
4. Wait for confirmation: "Ready! Loaded X chunks."

Step 3: Generate Summaries

1. Type your instruction in the chat input (examples below)
2. Press Enter
3. View the formatted summary in the chat area

Example Instructions

• "Summarize the main points in bullet format"
• "Teach this material to a Year 2 nursing student"
• "Extract financial risks and present in a table"
• "Give me a 3-paragraph executive summary"
• "Explain this like I'm 10 years old"

📁 Project Structure

summarization/
├── main.py              # Streamlit UI and chat interface
├── loaders.py           # Document loading logic for all formats
├── summarizer.py        # LLM and summarization chain logic
├── utils.py             # Logging utilities
├── pyproject.toml       # Project dependencies (uv/pip)
├── README.md            # This file
└── .venv/               # Virtual environment (auto-created)

🛠️ Technical Details

Dependencies

• langchain (0.3.14): LLM orchestration framework
• langchain-community (0.3.14): Community loaders and tools
• langchain-ollama: Ollama integration for LangChain
• langchain-text-splitters: Text chunking utilities
• streamlit: Web UI framework
• pypdf: PDF parsing
• python-docx: DOCX parsing
• python-pptx: PowerPoint parsing
• beautifulsoup4 + lxml: HTML parsing
• docx2txt: Additional DOCX support

How It Works

1. Document Loading (loaders.py):

   • Files are saved to a temporary directory
   • Appropriate loader selected based on file extension
   • Documents converted to LangChain Document objects

2. Text Splitting (summarizer.py):

   • Documents split into manageable chunks (default: 2000 chars, 200 overlap)
   • Preserves context across chunk boundaries

3. Summarization:

   • User instruction combined with formatting directive
   • Chain selected based on method (Stuff/Map Reduce/Refine)
   • LLM generates summary following instructions
   • Output extracted and formatted for display

4. Chat Interface (main.py):

   • Session state maintains document chunks and chat history
   • Real-time streaming and status updates
   • Clean Markdown rendering

🔧 Customization

Change the Default Model

Edit main.py or summarizer.py:

model_name = st.text_input("Ollama Model", value="llama3")  # Change here based on the ollama model you pulled

Adjust Chunk Size

Edit summarizer.py:

def split_documents(documents, chunk_size=3000, chunk_overlap=300):  # You can adjust here depending on your interest 

Modify Prompts

Edit the prompt templates in summarizer.py to customize how the LLM processes documents.

🐛 Troubleshooting

"No module named 'langchain.chains'"

• Ensure langchain==0.3.14 is installed (not 1.2.0)
• Run: uv sync or reinstall dependencies

"Connection refused to Ollama"

• Make sure Ollama is running: ollama serve
• Check Ollama is on default port 11434

"Model not found"

• Pull the model: ollama pull mistral-large-3:675b-cloud
• Or change to an available model in the sidebar

Slow summarization

• Use a smaller/faster model (e.g., mistral instead of mistral-large-3)
• Reduce chunk size or use "Stuff" method for smaller docs

📝 License

This project is provided as-is for educational and personal use.

🤝 Contributing

Contributions welcome! Feel free to:

• Report bugs
• Suggest features
• Submit pull requests

📧 Support

For issues or questions, please open an issue in the repository.

---

Built with ❤️ using LangChain, Streamlit, and Ollama