Rust-based PDF-to-Markdown extraction tool released

pdf-inspector, developed by the web crawling tool Firecrawl to optimize its own pipeline, analyzes PDF documents to determine in milliseconds whether they are text-based or scanned (image-based).

This library provides the following key features:

• PDF type classification: Without loading the entire document, it classifies documents into TextBased, Scanned, ImageBased, and Mixed types through sampling, and returns the page numbers without text so that OCR can be performed only on the necessary parts.
• High-performance Markdown conversion: Through font size, style, and layout analysis, it recognizes headings (H1-H4), lists, code blocks, and tables, extracting them in Markdown format.
• Efficient pipeline: Implemented in Rust, it enables local processing within 200ms, and shares parsing results between the classification and extraction stages to prevent duplicate I/O.

Benchmark results showed superior performance in table recognition performance and processing speed compared to existing tools such as pymupdf4llm.