opendataloader-pdf

Why we included this project

PDF-heavy document collections are a common pain point for retrieval pipelines, and this parser is built to sit at the extraction layer. It turns PDFs into Markdown, JSON with per-element bounding boxes, and HTML, keeping reading order and table structure intact so chunking and citation work cleanly downstream. A deterministic local mode handles straightforward documents quickly, while a hybrid mode sends complex pages, such as borderless tables, scanned files, formulas, and charts, to an AI backend, with built-in OCR for poor-quality scans. The accessibility angle is unusual: it can auto-tag untagged PDFs into screen-reader-ready Tagged PDFs, which matters if you need to comply with accessibility regulations. Python, Node.js, and Java SDKs, plus a LangChain integration, make it easy to fit into existing stacks.