Everything people actually use OCR for

The everyday one. Someone sent you a scan or a locked PDF and now you have to edit it, fill it in, translate it or reuse half of it. You do not want the text — you want the same document, editable.

• Columns, tables, indentation and images stay where they were
• No floating text boxes that explode the moment you type
• Fonts and spacing preserved, not substituted
• Works on photos of pages, not only clean scans

A large share of Persian conversion demand is not scans at all. It is digital PDFs made by older tools whose text turns into disconnected, meaningless letters the moment you copy it. People photograph a perfectly good page just to get readable text back.

• We read the page as printed rather than trusting a corrupted text layer
• Half-spaces reconstructed by measurement, not by guesswork
• Persian digits and punctuation preserved exactly
• Faster and cheaper than photographing pages one by one

Turn years of past papers into editable, re-orderable items — equations, geometry figures, option layout, answer keys and marking schemes intact — so they can be remixed into new papers and tagged by topic.

• Equations become real, editable Word equations
• Figures redrawn as vector shapes a teacher can adjust
• Question numbering reproduced exactly as printed
• Optional per-question JSON for an item bank

Engineering and science lecture notes are the single most expensive thing to retype, because typists charge separately for every formula and every table. This is where the manual cost is concentrated.

• Every formula compiled to prove it is valid before it ships
• Tables rebuilt as real tables, not pictures
• Mixed Persian and English handled in one pass
• Page boundaries kept so lecture references still work

Financial statements, transcripts, price lists, laboratory results, census sheets — the table is the whole point, and it is exactly what general OCR flattens.

• Merged cells and multi-page tables reconstructed
• Original numerals kept — Persian, Arabic-Indic or Latin
• Column alignment used as evidence, not decoration
• Per-cell confidence so you know what to check

An official translation has to mirror the original page: same stamps, same table of grades, same signature blocks. Translators currently rebuild that layout by hand before they can begin translating.

• A visual replica with editable text runs, ready to translate in place
• Dense grade tables with Persian numerals reproduced exactly
• Seals and logos placed as clean cut-outs
• Page-for-page correspondence for certification

In these fields the page number is not decoration, it is a condition of validity: citations are made as volume and page. A text that loses its page boundaries is worthless to the researcher who needs to cite it.

• Volume, page and line preserved as citable coordinates
• Diacritics reproduced exactly — never added, never dropped
• Arabic and Persian letterforms never silently converted
• Marginal notes kept as linked annotations, not merged into the body

Handwritten material costs ten to twenty times what printed material costs to transcribe, and there is no serious commercial tool for Persian or Arabic manuscript hands. This is the widest gap in the whole market.

• Two independent readings with every disagreement surfaced
• Uncertain regions flagged for a human rather than invented
• Line-level coordinates back to the original image
• Critical-apparatus output when you supply two witnesses

Give us two scans of the same text and get their differences as a critical apparatus. Existing collation tools are built for Latin scripts and tokenise Arabic script badly.

• Word-level differences with variant sigla
• Arabic-script-aware tokenisation and normalisation
• Output as an editable apparatus, not a diff dump
• Every variant links back to both source images

An out-of-print book exists only as paper or as a flat scan. To reissue it you need the page design back in an editable form — which is exactly what typesetting vendors charge several dollars a page to rebuild.

• Page design preserved for print, structure preserved for ebook
• Figures redrawn as vectors so they print sharp
• Fidelity score per page, so you know where to look
• Far below the per-page cost of a typesetting contractor

This is the one job where page fidelity is the wrong answer. Blind readers need clean reflowable text with correct language tags — and in Persian a misplaced half-space makes a screen reader mispronounce the word.

• Reflowable structure with real headings and reading order
• Mathematics as MathML, not as a picture with alt text
• Persian spacing corrected for pronunciation, not for looks
• Deadline-driven: the rules already apply in the EU

Retrieval systems are now the largest new buyer of layout-faithful reading, because a chatbot that cannot say which page it answered from is not usable in a regulated setting.

• Clean Markdown with a real document tree
• Bounding boxes and page anchors on every block
• Tables as HTML so they survive chunking
• Citations that resolve back to the exact page region

Courts increasingly require a genuine searchable text layer for e-filing, and a missed word in discovery is a missed document. Separately, taking the other side's scanned contract into Word to redline is its own daily job.

• Searchable PDF with an accurate invisible text layer
• Original scan untouched underneath
• Scanned contracts into clean, redlinable Word
• Page-anchored text so you can point at a line

Organisations with rooms full of paper need it searchable and retrievable, with an audit trail — not converted into something new.

• Batch processing of whole shelves, resumable
• Searchable PDF plus library-standard ALTO and METS
• Runs inside your own network when documents may not leave
• Per-page verification report for the audit file

Two-column layout, footnotes, running heads and multi-page result tables are where ordinary extraction quietly falls apart — and where researchers still copy numbers by hand.

• Reading order correct across columns and footnotes
• Result tables extracted as tables, ready for analysis
• Equations as LaTeX or as editable Word equations
• Reference lists kept intact and parsable

Dense multi-column pages where one article wraps around three others. Article zoning is what makes the archive searchable rather than merely scanned.

• Article-level zoning, not just page-level text
• Headlines, bylines and captions labelled
• Library-standard ALTO and METS output
• Bulk text for research APIs

Typing offices are not only competitors — they are a distribution channel. They already have the customers and the deadlines; what they lack is anything that handles tables and formulas without a human.

• API and batch upload for whole jobs
• Your own accumulated corrections become your private rules
• White-label output with no branding
• Volume pricing, invoiced monthly

Organisations constantly need last year's tender document, form or standard operating procedure back in editable shape — with the boxes, rules and numbering exactly as they were.

• Boxes, rules and shaded fields reproduced
• Numbering schemes kept exactly as printed
• Optionally converted into fillable fields
• Consistent house style across a whole department

Registry entries, endowment deeds, old letters and family papers — usually handwritten, often the only copy, and the reason someone finally scans a box in the attic.

• Handwriting handled with uncertainty made visible
• Original image kept beside every transcription
• Searchable across a whole collection
• Nothing invented where the page is illegible

A three-hundred-page book uploaded once, processed page by page, with a live view of which pages are done and which need a second look.

• Resumable — a dropped connection does not restart the book
• Per-page fidelity scores and a queue of flagged pages
• Reviewer assignment for the pages that need eyes
• One certificate of conformance for the whole job