DocuAgent — AI Solution to Structure Your Documents

DocuAgent is an intelligent document-processing pipeline that leverages advanced AI models to automatically extract, structure, and summarize content from PDF documents.

Overview

DocuAgent combines state-of-the-art computer vision, deep learning, and natural language processing techniques to transform unstructured PDFs into structured, machine-readable formats.

Whether you're processing financial reports, invoices, technical papers, or complex handwritten forms, DocuAgent can:

• Extract text from documents
• Detect layout regions and document structure
• Recognize and convert tables to structured formats
• Produce structured outputs in JSON format
• Generate concise summaries

Features

• PDF Processing – Convert PDFs to images and extract text
• Layout Recognition – Detect headers, paragraphs, tables, figures, and other document elements
• Table Detection & Extraction – Parse complex tables into structured formats
• Flexible Output – JSON format with layout, table structure, and content
• Configurable Pipeline – YAML/JSON-based configuration system
• FastAPI Backend – Real-time document processing and serving

Project Structure

docuagent/
├── data/                          # PDFs, datasets (train/val/test splits)
├── src/                           # Source code
│   ├── preprocessing/             # PDF to image, OCR, text extraction
│   ├── models/                    # LayoutLM, TableFormer, summarizers
│   ├── utils/                     # Helper functions (logging, configs)
│   └── pipeline.py                # Main pipeline orchestration script
├── notebooks/                     # Jupyter/Colab notebooks for experiments
├── outputs/                       # Structured results (JSON, summaries, charts)
├── requirements.txt               # Project dependencies
├── README.md                      # This file
└── .gitignore                     # Git ignore rules (large files, venv, cache)

Installation

Prerequisites

• Python 3.8 or higher
• pip or conda package manager
• For GPU acceleration: CUDA 11.0 or higher

Setup Instructions

1. Clone the Repository

git clone <repository-url>
cd docuagent

2. Create a Virtual Environment

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

3. Install Dependencies

The installation process requires two steps due to PaddlePaddle dependencies.

Step 1: Install PaddlePaddle with GPU Support

Visit the official PaddlePaddle installation page and select the version compatible with your CUDA installation:

https://www.paddlepaddle.org.cn/en/

Follow the installation command provided for your specific CUDA version.

Step 2: Install Remaining Dependencies

Before proceeding, verify your PaddlePaddle installation is compatible with PaddleOCR. Refer to the compatibility table:

• PaddleOCR 3.0.0 requires PaddleX 3.0.0 and PaddlePaddle ≥ 3.0.0
• PaddleOCR 3.1.x requires PaddleX ≥ 3.1.0, < 3.2.0 and PaddlePaddle ≥ 3.0.0

Full compatibility documentation: https://www.paddleocr.ai/main/en/version3.x/paddleocr_and_paddlex.html

Once PaddlePaddle is installed and verified, install the remaining dependencies:

pip install -r requirements/base.txt

Demo

Watch the demonstration video: https://youtu.be/nZaeeclYW-0

Usage

Documentation for running the pipeline and API usage coming soon.