Document Extractor (docxtract)

A CLI tool for extracting structured content from academic PDF documents and generating high-quality Chinese summaries.

Features

• 📄 Extract text content from PDF files
• 🔍 Automatically detect and parse academic paper sections
• 🇹🇼 Generate Chinese summaries using GPT-4.1 (per-section summarization, with specialized prompts)
• 📝 Output structured content as Markdown files (including per-section and overall summary)
• 🚀 Simple CLI interface with rich progress indicators

Installation

1. Clone the repository:

git clone <repository-url>
cd llm-document-extracter

2. Install dependencies using uv:

uv sync

Usage

Generate Chinese Summary (Markdown Output)

To generate high-quality Chinese summaries, you need to configure Azure OpenAI:

1. Copy the environment template:

cp .env.example .env

2. Edit .env and add your Azure OpenAI credentials:

AZURE_OPENAI_ENDPOINT=https://your-resource-name.openai.azure.com/
AZURE_OPENAI_API_KEY=your-api-key-here
AZURE_OPENAI_API_VERSION=2024-02-15-preview
AZURE_OPENAI_DEPLOYMENT_NAME=your-gpt4-deployment-name

Command Options

python main.py summary [OPTIONS] PDF_PATH

Options:
  -o, --output PATH    Output path for the Markdown file
  -p, --preview        Preview the output without saving to file
  --help              Show help message

Examples

# Generate summary and save to default output path (/summaries/**.md)
python main.py summary research_paper.pdf

# Generate summary with custom output path
python main.py summary research_paper.pdf -o summaries/paper.md

# Preview summary markdown without saving
python main.py summary research_paper.pdf --preview

Output Format

The tool generates Markdown files with the following structure:

# Paper Title

## Top-5 Important Points

1. **Important Point 1**
   Description of the first key finding or contribution

2. **Important Point 2**
   Description of the second key finding or contribution

3. **Important Point 3**
   Description of the third key finding or contribution

4. **Important Point 4**
   Description of the fourth key finding or contribution

5. **Important Point 5**
   Description of the fifth key finding or contribution

## Application Ideas

1. **Application Area 1**
   Potential application or future research direction

2. **Application Area 2**
   Another potential application or research direction

3. **Application Area 3**
   Additional application possibilities

## Chinese Summary

(Overall summary generated from all section summaries in Traditional Chinese)

---
*Chinese summary generated using GPT-4.1*

Project Structure

docxtract/
├── cli.py          # CLI interface using Typer
├── extract.py      # PDF text extraction with PyMuPDF
├── parser.py       # Section header detection and parsing
├── summarizer.py   # Chinese summary generation with Azure OpenAI (per-section logic)
├── chain.py        # LangChain LCEL pipeline for document processing
├── writer.py       # Markdown file output
├── models.py       # Pydantic data models
└── utils.py        # Shared utilities

Useful CLI Commands

• make sync     # Initialize project environment (install dependencies, setup venv)
• make dev     # Run server locally (FastAPI, hot reload)
• make lint     # Check code lint using ruff
• make format     # Format code using ruff
• make check     # Check code typing using pyright
• make clean     # Clean Python cache files
• make unit-test     # Run unit tests

Development

Code Quality

The project follows these standards:

• Type hints for all functions
• Pydantic v2 for data validation
• Single Responsibility Principle (SRP)
• Clear English comments
• Comprehensive error handling

Requirements

• Python 3.12+
• PyMuPDF for PDF parsing
• Typer for CLI interface
• Pydantic v2 for data validation
• LangChain OpenAI for summary generation
• Rich for beautiful CLI output

Future Development

• Phase 2: RAG storage and retrieval
• Phase 3: Personal knowledge base with LLM integration
• Batch processing for multiple PDFs
• Enhanced section detection algorithms
• Support for different document types

License

See LICENSE file for details.