Inspiration
On the first day of my math class in spring quarter, my professor walked in with bad news. He could no longer post his handwritten lecture notes on Canvas. The reason was a new federal rule called ADA Title II, which requires every page of online course content to meet WCAG 2.2 AA standards by April 2027. For text, simple. For matrix proofs and integrals that's two hours of manual LaTeX typing per lecture hour. So he just stopped posting altogether. That moment stuck with me. Twenty-three of us were going to study from whatever we caught in real time. Blind and low-vision STEM students were already locked out of higher math entirely. And my professor was being punished for teaching a visual subject. The whole thing felt solvable, so I started building.
What it does
AccessNotes converts handwritten STEM lecture notes into WCAG 2.2 AA compliant digital materials. A professor uploads a scanned page. Our application recognizes equations, parses matrix structure, and outputs MathML inside accessible HTML that any screen reader can read aloud. A side-by-side editor lets professors verify and tweak the output in seconds. Two hours of manual LaTeX transcription becomes two minutes.
How we built it
A Next.js frontend handles uploads and the side-by-side editor. A FastAPI backend pipes the scan into Google Gemini's multimodal model, which identifies layout regions, recognizes handwritten equations, and reconstructs the mathematical structure. Output is wrapped in MathML and WCAG 2.2 AA compliant HTML with proper ARIA labels. KaTeX renders the live preview. Hosted on Vercel and Railway.
Challenges we ran into
Multi-line proofs that span four or five physical rows had to be reconstructed as one logical chain, not disconnected equations. Handwritten matrices have inconsistent spacing, so we built a dedicated detection pass before the main recognizer. Screen readers also disagree with each other. NVDA, JAWS, and VoiceOver each phrase the same equation differently, so we tested across all three.
Accomplishments that we're proud of
We built a working pipeline that takes a real scanned page from a UC Davis math class and returns screen-reader-ready HTML in under a minute. We solved the multi-line expression problem most existing OCR tools fail on. And we did it during a single quarter, against a federal compliance deadline that nobody else has built specifically for STEM.
What we learned
MathML, not LaTeX, is the actual accessibility standard. Screen readers handle math differently from each other, so testing across NVDA, JAWS, and VoiceOver is non-negotiable. And the gap between "AI can read text" and "AI can read handwritten calculus correctly" is much wider than we expected.
What's next for AccessNotes
Pilot with the UC Davis math department this fall, then expand across STEM departments at UC and CSU campuses. Add support for chemistry notation, physics diagrams, and engineering schematics. Build an LMS integration for Canvas so professors can upload directly from their existing workflow. Push toward the April 2027 federal deadline as the default STEM compliance tool in higher education.
Log in or sign up for Devpost to join the conversation.