Arabic OCR Expert: Fix Layout & Accuracy Issues in Legal Documents

We run a legal AI platform on Arabic official gazette documents. Our current OCR (Azure) makes systematic errors that we need to fix. Use any approach that works: existing APIs, vision LLMs, a custom-trained model, or a hybrid. We care about results, not method. Problems to solve (see attached screenshots) Lines with dotted placeholders ("......") come out in the wrong order (RTL scrambled) Misread characters (Persian letters instead of Arabic, "«" read as "((") Missed, reversed, or broken tables Reversed compound article numbers (e.g. 430-1 vs 1-430) Deliverable A working, documented solution (Python preferred) that we can plug into our weekly pipeline and run on our existing archive, with measurable accuracy on a test set. Requirements Proven Arabic OCR experience (RTL, tables, mixed Arabic/numerals) Must read Arabic Legal/official document experience is a plus To apply Please review the attached screenshots first, then include in your proposal: How you would handle the cases shown (3-5 lines, specific to the screenshots) Relevant Arabic OCR work Your timeline Proposals that don't address the screenshots will not be considered.