Lab · Exploring
Accessible PDFs, with AI
Forms, bills, medical reports, contracts. The documents that matter most often arrive as PDFs, and most of them are hard or impossible to read with a screen reader.
The problem
An accessible PDF needs structure under the surface: headings, reading order, tables that make sense, text alternatives for images, the right language. Fixing a long document by hand takes hours, and it needs skills most teams don't have.
With the European Accessibility Act, more and more services have to get this right. Today many of them simply can't keep up.
The question I'm exploring
Can AI do the heavy lifting, with a person checking the result? AI is good at guessing structure and describing images. It's not good at being sure. So the interesting design work is in between:
- what the tool fixes on its own, and what it asks a person to confirm;
- how it shows what it changed, so a non-expert can trust it;
- how it checks the result against WCAG and PDF/UA, instead of just saying "done".
Where I am
It's a personal project I work on in the evenings and at weekends. Early days, so no screens yet: I'll share more when there's something worth showing.
Do you deal with inaccessible PDFs at work? Tell me about it: real problems are the best input I can get.