opendataloader-project/opendataloader-pdf ? reverse-engineered prompt

I need a tool to process PDF files for my RAG pipeline. It has to be really good at understanding document structure so I can get clean data for my AI models.

The main thing is accurately extracting content into structured formats like Markdown and JSON. It should handle complex layouts, like multi-column articles or scientific papers, and keep the correct reading order. It must also be great at pulling out tables, even ones without clear borders. For every single element—headings, paragraphs, list items—I need the bounding box coordinates included in the output so I can cite the source location in the original PDF.

The tool should also work with scanned documents, so it needs a built-in OCR capability. Finally, I want it to help with accessibility compliance by automatically analyzing a PDF's layout and converting it into a tagged PDF that's screen-reader friendly. Build the core engine in Java, but please provide easy-to-use SDKs for Python and Node.js.