Cascadeflagship

A multi-engine OCR pipeline for PDF documents with per-region routing, hand-edit detection, redaction marking, and a side-by-side web viewer. Cascade runs each page through an ordered list of OCR engines — cheapest first, heaviest last — and stops at the first engine that clears a confidence threshold. Regions mode takes it further: Docling's layout model identifies blocks on the page (printed text, handwritten text, tables, pictures), and each block is routed through its own configured engine ladder. Mixed pages — a printed letterhead with a handwritten body, a court filing with margin annotations — only pay for the engine each piece actually needs.