Startups Tackle PDF Parsing For Document Search

Since last November, technologists and researchers have advanced specialized PDF-parsing tools to extract structured data from millions of government and archival PDFs, including 20,000 House Oversight pages and more than three million DOJ files. Companies like Reducto and research teams at the Allen Institute developed vision-language models (e.g., olmOCR) and datasets to improve OCR, table parsing, and document understanding, promising faster, searchable access to formerly unusable document corpora for investigators and practitioners.