4 ms·

Show HN: I improved my handwritten math OCR (now preserves derivations)

1 points by mrajatnath 8mo ago

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 https://www.useaxiomnotes.com/app Happy to answer questions.

mrajatnath 8mo ago
Rajat here. A few technical details about how this works. Stack: - Next.js - Tailwind - KaTeX for rendering - Supabase storage - deployed on Vercel The pipeline is roughly: image → vision model → Markdown + LaTeX → custom renderer The tricky part isn’t OCR itself — it's preserving structure. Examples: • consecutive equations with aligned `=` signs need to become a single `align` block • handwritten tables must be reconstructed from vertical alignment patterns • numbered problems must stay separate instead of merging The system prompt ended up being ~300 lines mostly consisting of *negative constraints* like: - don't simplify math - don't merge derivation steps - don't reorder columns Without those rules the model constantly tries to "improve" the notes. One surprising lesson: prompt engineering for OCR is very different from chat prompts — you want the model to be extremely literal. Still working on better handling for diagrams and messy annotations. Curious if anyone here has worked on *math layout detection or document AI*.
gus_massa 8mo ago
Do you have a few samples? Is it possible to use it like detextify, writing in a canvas in the browser?
- mrajatnath 8mo ago
  Right now the workflow is based on images rather than drawing input. You upload a photo or PDF of handwritten notes, and Axiom tries to recover the mathematical structure from that page — things like aligned equations, multi-step derivations, numbered problems, and tables — and return it as structured Markdown + LaTeX. Most OCR tools already recognize the characters reasonably well, but they tend to lose the layout that makes math readable (for example aligned `=` signs or derivation steps). The main focus here is preserving that structure. I’m planning to add a few example pages on the site so people can see typical input → output results. Regarding something like Detexify: not yet. Detexify is more about recognizing individual symbols drawn in a canvas, while Axiom currently works on full handwritten pages. That said, a browser canvas mode for sketching equations is something I’ve been thinking about for the future, especially for quick LaTeX generation.
  - gus_massa 8mo ago
    Yes, I was wondering about a Detextify for equations, like \int_2^7 x sin(x+1) dx

1 points by mrajatnath 8mo ago

mrajatnath 8mo ago

Rajat here. A few technical details about how this works. Stack: - Next.js - Tailwind - KaTeX for rendering - Supabase storage - deployed on Vercel The pipeline is roughly: image → vision model → Markdown + LaTeX → custom renderer The tricky part isn’t OCR itself — it's preserving structure. Examples: • consecutive equations with aligned `=` signs need to become a single `align` block • handwritten tables must be reconstructed from vertical alignment patterns • numbered problems must stay separate instead of merging The system prompt ended up being ~300 lines mostly consisting of *negative constraints* like: - don't simplify math - don't merge derivation steps - don't reorder columns Without those rules the model constantly tries to "improve" the notes. One surprising lesson: prompt engineering for OCR is very different from chat prompts — you want the model to be extremely literal. Still working on better handling for diagrams and messy annotations. Curious if anyone here has worked on *math layout detection or document AI*.

gus_massa 8mo ago

Do you have a few samples? Is it possible to use it like detextify, writing in a canvas in the browser?

mrajatnath 8mo ago
Right now the workflow is based on images rather than drawing input. You upload a photo or PDF of handwritten notes, and Axiom tries to recover the mathematical structure from that page — things like aligned equations, multi-step derivations, numbered problems, and tables — and return it as structured Markdown + LaTeX. Most OCR tools already recognize the characters reasonably well, but they tend to lose the layout that makes math readable (for example aligned `=` signs or derivation steps). The main focus here is preserving that structure. I’m planning to add a few example pages on the site so people can see typical input → output results. Regarding something like Detexify: not yet. Detexify is more about recognizing individual symbols drawn in a canvas, while Axiom currently works on full handwritten pages. That said, a browser canvas mode for sketching equations is something I’ve been thinking about for the future, especially for quick LaTeX generation.
- gus_massa 8mo ago
  Yes, I was wondering about a Detextify for equations, like \int_2^7 x sin(x+1) dx