opendataloader-pdf: Extract Structured Data and Add Accessibility Tags to PDFs

Overview

OpenDataLoader PDF is a PDF parser and accessibility tool. It extracts document content into Markdown, JSON, and HTML, preserving layout information such as reading order and element coordinates for downstream applications.

It also generates Tagged PDFs from untagged files, giving teams a way to automate part of PDF accessibility remediation. The core project is Apache-2.0 licensed; PDF/UA export and a visual accessibility editor are described as enterprise features.

Key Features

• Convert PDFs to Markdown, JSON, HTML, text, and annotated PDF output.
• Include page numbers, semantic element types, and bounding boxes in structured output.
• Extract headings, lists, tables, images, and document reading order.
• Use an optional hybrid mode for complex layouts, OCR, formula extraction, and image descriptions.
• Generate Tagged PDFs from untagged documents.
• Process files through Java, Python, or Node.js interfaces, with a command-line option.
• Filter certain hidden or suspicious PDF text to reduce prompt-injection risks in AI workflows.

Use Cases

• RAG developers can preserve headings, tables, and page coordinates when preparing PDF content for retrieval and citations.
• Document-processing teams can batch-extract structured data from digital PDFs without building a parser from scratch.
• Accessibility teams can automatically generate Tagged PDFs as an initial remediation step for untagged documents.
• Teams handling scanned or layout-heavy PDFs can try hybrid processing with OCR when local extraction is insufficient.

Getting Started

For Python, install the package and convert a PDF:

pip install -U opendataloader-pdf

import opendataloader_pdf

opendataloader_pdf.convert(
    input_path="file.pdf",
    output_dir="output/",
    format="markdown,json"
)

See the README for Java and Node.js setup, hybrid mode, and additional options.

Alternatives

• text-extract-api: It exposes document extraction as a self-hostable API using OCR and Ollama, while OpenDataLoader PDF offers Java, Python, and Node.js SDKs and hybrid processing.
• unstructured: It parses many document types into structured elements through a Python library, while OpenDataLoader PDF focuses on PDFs and also supports PDF auto-tagging.
• marker: It converts PDFs and other documents to Markdown, JSON, HTML, or chunks using local models, while OpenDataLoader PDF offers multiple SDKs and hybrid processing.
• docling: It handles PDFs alongside Office, image, and audio formats, while OpenDataLoader PDF focuses on PDFs and includes automatic tagging for accessibility.

Considerations

The Python quick start requires Java 11 or later in addition to Python. The Node.js package requires Node.js 22.13 or later. Hybrid processing adds a separately installed backend and may require additional OCR engines or language data depending on the setup. The project reports regular commits and releases and ongoing issue and pull-request activity, but benchmark figures in the README are project-published results and may not reflect every document collection. The README also distinguishes free Tagged PDF generation from enterprise PDF/UA export and visual editing.