OpenDataLoader-PDF: An open source tool for structured PDF parsing (github.com)

OpenDataLoader-PDF is an open‑source, local‑first PDF parsing tool designed to turn PDFs into structured JSON, Markdown or HTML that’s ready for LLMs, vector search and RAG pipelines. It reconstructs document layout — headings, lists, tables (including borderless/merged cells via an optional Table AI), images and reading order — so content can be chunked, indexed and queried more reliably. The tool is rule‑based and high‑throughput (no GPU required), offers OCR for scanned pages, produces annotated PDFs visualizing detected structure, and enables AI‑safety by default by filtering likely prompt‑injection content. The project publishes transparent performance and adversarial (red‑teaming) benchmarks, signaling maturity for production ingestion workflows. On the integration side, OpenDataLoader‑PDF is a Java CLI core (requires Java 11+) with wrappers for Python (pip install opendataloader-pdf), Node/npm (@opendataloader/pdf) and Docker; it also provides Maven/Gradle examples and a Java API. Outputs include a rich JSON schema with nodes (type, bounding box, fonts, page number, table rows/cells, etc.), plus options to preserve line breaks, include images or HTML in Markdown, replace invalid characters, and selectively disable safety filters. Because it runs entirely locally and exposes detailed layout metadata, it’s immediately useful for ML engineers building reliable ingestion, indexing, and safety‑aware RAG systems.