Why equations break normal converters
Fractions, subscripts, summations, matrices — a PDF stores all of that as drawn glyphs, so plain text extraction returns unreadable soup. The Math & Scientific Document OCR card recognizes formulas visually with the Granite-Docling model and re-emits them as standard LaTeX: $f(x)$ inline, display equations as blocks, exactly what Overleaf, Jupyter, and Word's equation editor expect.
More than formulas
Scientific papers are rarely math alone, so the card also keeps programming code inside fenced code blocks and preserves the surrounding headings and paragraphs. Export Markdown with LaTeX for notes and papers, Word with formatted equations for collaborators, JSON formula nodes for processing, or plain text.
It works on both clean digital PDFs and photographed lecture notes, and it supports the same nine recognition languages as the rest of the suite. Need other formats or mixed content? The same engine backs the universal OCR suite — one workflow for the whole paper, formulas included.
Workflow tips that save hours
When a page is noisy, crop the scan down to just the equation region — less clutter means higher recognition confidence. The LaTeX the card emits compiles as-is in Overleaf, renders in MathJax and KaTeX, and pastes cleanly into Word's equation editor, so the output drops straight into whatever you were already using. Equations with a code companion (think computational notebooks) stay inside fenced code blocks, while surrounding prose keeps its heading structure. And for full papers where you want text, formulas, and code in one continuous file, pair this card with AI-Ready Markdown and let the two outputs complement each other.
One more tip for textbooks: dense two-column layouts occasionally reorder formula fragments, so if an equation looks scrambled, crop tighter around the expression itself. The LaTeX re-renders exactly as recognized, which makes eyeballing the output against the original a ten-second check that catches anything the model second-guessed.