If you've ever tried to extract text from a scanned Arabic document, you already know the pain. Most OCR tooling is built English-first. Arabic adds three problems on top: Right-to-left (RTL) text that breaks naive layout assumptions. Connected letters (ligatures) — the same letter changes shape depending on its position in the word. Diacritics and a different numeral set that generic models drop or mangle. The result: you run a scanned Arabic contract, invoice, or governm...