PDF to LaTeX Outliner

Generate LaTeX outlines from PDF documents using local AI models. No API keys required!

Features

• Runs completely locally using Ollama
• Extracts text from PDFs intelligently
• Generates LaTeX outlines based on your examples
• Available as both Python script and Docker container
• Private - your documents never leave your machine

Quick Start (Python - Recommended)

Prerequisites

1. Install Ollama: ollama.com

   # Mac/Linux
   curl -fsSL https://ollama.com/install.sh | sh
   
   # Windows: Download from ollama.com

2. Install Python 3.11+: python.org

Installation

# Clone or download this repository
git clone https://github.com/PrismaticFlame/LaTeXOutlineGenerator
cd pdf-outliner

# Install Python dependencies
pip install -r requirements.txt

# Pull the AI model (one-time, ~2GB download)
ollama pull llama3.2:3b

Usage

1. Add your example outlines to the examples/ directory as .tex files
2. Place your PDF in the input/ directory
3. Run the script:

# Basic usage
python outline.py input/your-document.pdf

# Specify custom output location
python outline.py input/your-document.pdf -o my-outline.tex

# Use a different model
python outline.py input/your-document.pdf -m llama3.2:1b

Your outline will be generated in output/your-document_outline.tex

Docker Usage

For those who prefer containers or want a completely isolated environment.

Prerequisites

• Docker installed (docker.com)

Build the Image

docker build -t pdf-outliner .

Run with Docker

# Make sure your directories exist
mkdir -p input output examples

# Process a PDF
docker run --rm \
  -v $(pwd)/input:/app/input \
  -v $(pwd)/output:/app/output \
  -v $(pwd)/examples:/app/examples \
  pdf-outliner input/your-document.pdf

Even Easier: Docker Compose

# Build the image
docker-compose build

# Process a PDF
docker-compose run pdf-outliner input/your-document.pdf

Example Outlines (Important!)

The AI learns best from paired examples: a PDF and its corresponding LaTeX outline.

Example Directory Structure

examples/
├── example1.pdf              # Your first sample document
├── example1_outline.tex      # The outline you want for this type of document
├── example2.pdf              # Another sample document (different style/topic)
└── example2_outline.tex      # Its corresponding outline

Naming Convention

• PDFs: example1.pdf, example2.pdf, etc.
• Outlines: example1_outline.tex or example1.tex (both work)

The script automatically pairs them by matching the base name.

What the AI Learns

When you provide paired examples, the AI learns:

• Structure patterns: How you organize sections and subsections
• Depth preferences: Do you use subsubsections? How deep do you go?
• Style choices: Formal vs informal section naming
• Content mapping: What parts of the PDF become what sections

Example outline (examples/example1_outline.tex):

\documentclass{article}
\begin{document}

\section{Introduction}
\subsection{Background}
\subsection{Motivation}

\section{Main Content}
\subsection{Key Concept 1}
\subsection{Key Concept 2}

\section{Conclusion}

\end{document}

Can I Use Only .tex Files?

Yes! If you don't have sample PDFs, you can just provide .tex outlines. The AI will still learn the LaTeX structure, but won't understand as well when to apply which patterns. It works, just not as effectively.

More examples = better results! Start with 2-3 paired examples for best quality.

Project Structure

pdf-outliner/
├── outline.py              # Main Python script
├── requirements.txt        # Python dependencies
├── Dockerfile             # Docker image definition
├── docker-compose.yml     # Easy Docker setup
├── docker-entrypoint.sh   # Docker startup script
├── examples/              # Your example .tex files
│   ├── example1_outline.tex
│   └── example2_outline.tex
├── input/                 # Place PDFs here
└── output/                # Generated outlines appear here

Available Models

You can use different Ollama models depending on your needs:

• llama3.2:1b - Fastest, lowest resource usage (~1GB RAM)
• llama3.2:3b - Recommended - Good balance (~4GB RAM)
• llama3.2:7b - Better quality, slower (~8GB RAM)
• mistral:7b - Alternative, good quality (~8GB RAM)

Change the model with the -m flag:

python outline.py input/doc.pdf -m llama3.2:1b

Troubleshooting

"Ollama connection refused"

Make sure Ollama is running:

ollama serve

"Model not found"

Pull the model first:

ollama pull llama3.2:3b

Docker: Model downloads every time

Uncomment the model download line in the Dockerfile to pre-download the model into the image (increases image size by ~2GB).

Poor outline quality

• Add more example outlines in the examples/ directory
• Try a larger model (llama3.2:7b or mistral:7b)
• Ensure your example outlines are consistent in style

System Requirements

Python Version:

• CPU only: 4-8GB RAM (for 3B model)
• GPU (optional): Faster processing

Docker Version:

• 8GB+ RAM recommended
• 10GB disk space (for model + container)

Contributing

Contributions welcome! Feel free to:

• Add support for more document formats
• Improve prompt engineering
• Add more model options
• Improve error handling

License

MIT License - feel free to use for personal or commercial projects.

Acknowledgments

• Ollama - Local AI model runtime
• pymupdf4llm - PDF text extraction
• Llama models by Meta

---

Questions? Open an issue or check the Ollama documentation