No, you couldn't do this before an LLM because if you could it would have been done already
I just did the impossible and you can too
I’m tired because I spent all day building a Claude Code skill to remediate handwritten math notes in Canvas courses. And I came up with a solution that I have not seen in all of the technology out there. Which is notable, because there is a compliance deadline fast approaching to get courses fully accessible (WCAG 2.1 under ADA Final Rule), which affects all public universities, has devastating penalties, and has spurred a frenzy of edtech spending on accessibility software. And yet, here is a solution I have not seen.
It’s one of 8 interrelated Claude Code skills I’ve been building to explore how we might use LLMs to address accessibility concerns. I’m tired as I narrate this video from working all day on this, but what I want you to know is that there isn’t, to my knowledge, any product on the market that does what you are about to see, and if it had been possible to do with non-LLM software there absolutely would have been.
Will it make me rich? No, because while I may have facility with these agentic systems there’s an abundance of people who can do this. If I discovered this this week I’m sure someone else figured this out three weeks ago. But they wouldn’t have figured it out a year ago because a year ago the base capabilities and structures to do this did not exist. I won’t be rich, but we will be able to build better and more accessible systems because of this, and that is a very good thing.
I say this because every time you show something to someone and say this is some new thing these systems do that could not have been done before people claim they could have built it with twine, 10 lines of Perl, and a toothpick. So here’s something you can do now that you know was impossible before — because if it had been possible it would have been done. Enjoy.
Update:
This is insanity (full pass of all 8 skills on a Canvas Commons Calculus course). Have to check the work, but if it does as well as it has done in my medium sized tests…
The AP Calculus AB remediation is fully complete. Here’s a summary of what was accomplished:
Remediation Results:
- 88 HTML pages: Added lang=”en”, <h1> headings, fixed heading hierarchy, replaced bare URL links, titled iframes
- 64 PDFs: Added /Lang en and dc:title metadata
- 3 DOCX: Added title, language, and Heading 1
- 49 PPTX (converted from legacy .ppt): 2,149 images described with alt text (100%), 886 complex images with full long descriptions in slide notes across 561 slides
- 6 QTI assessments: Text-only, no remediation needed
Deliverables ready at:
- queue/output/ap-calculus-ab/ap-calculus-ab-remediated.imscc (53.2 MB)
- queue/output/ap-calculus-ab/pptx-for-review/ — 49 PPTX files + review report for instructor verification
- queue/output/ap-calculus-ab/remediation-report.md — full report with instructor action itemsInstructor action items:
1. Review PPTX image descriptions in pptx-for-review/ (say “reintegrate pptx” when done)
2. Verify video captions on YouTube/Vimeo/Educreations embeds
3. Add alt text to 7 images in L’Hôpital’s Rule DOCX manually
4. Upload remediated IMSCC to Canvas

How does almost any knowledge work stay defensible in this new world? Yes, I get SaaS, workflows, APIs, proprietary databases, etc. But this kind of augmentation and automation work, how does any intellectual moat remain?
Mike, are you planning on sharing the skills files that you've developed for this workflow? I've got a very similar use case for transcribing handwritten pdfs of math notes into Canvas wiki pages with LaTeX.