Back to browse
I improved my handwritten math OCR (now preserves derivations)

I improved my handwritten math OCR (now preserves derivations)

by mrajatnath·Mar 4, 2026·1 point·4 comments

AI Analysis

●●●BangerSolve My ProblemEye CandyShip It

Math OCR that preserves derivation structure, not just transcription—solves a real student pain.

Strengths
  • Genuine problem: existing OCR destroys math structure; Axiom focuses on layout preservation, not character accuracy alone.
  • Clean web interface with upload, preview, and export—feels like a real product, not a proof-of-concept.
  • LaTeX + Markdown output is searchable and future-proof, not locked into proprietary format.
Weaknesses
  • No details on accuracy metrics, model choice, or how layout preservation actually works under the hood.
  • Math OCR is a narrow use case; Mathpix and Adobe Acrobat already handle this, though may not preserve structure equally well.
Category
Target Audience

Students, researchers, academics digitizing handwritten STEM notes

Similar To

Mathpix · Adobe Acrobat OCR · Snip & Sketch

Post Description

I built this after almost losing a semester’s worth of handwritten math derivations.

I was taking a Signals and Systems course and filling notebooks with Laplace transforms and long derivations. Before finals I tried digitizing them so I could search my notes.

Everything failed.

Most OCR tools can recognize the characters, but they destroy the structure that makes math readable:

- aligned equations lose alignment - multi-step derivations collapse into paragraphs - numbered problems merge together - tables flatten into plain text

So I built *Axiom*.

Instead of focusing only on transcription accuracy, it focuses on *preserving mathematical structure*.

Upload a photo of handwritten STEM notes and it returns structured Markdown with real LaTeX — keeping aligned equations, derivation steps, and problem blocks intact.

Under the hood it’s basically:

image → vision model → structured Markdown + LaTeX → KaTeX render

Most of the work ended up being in *layout preservation*, not OCR.

https://www.useaxiomnotes.com/app

Happy to answer questions.

Similar Projects

Security●●Solid

A ground up TLS 1.3 client written in C

Educational TLS 1.3 implementation, useful for learning but production alternatives exist.

WizardryShip It
theotrama
303mo ago