Skip to content

DocEngine now drafts accessibility templates with AI

Published:
2 min read

Most accessibility vendors open with a demo. Ours opens with your PDF. Upload one sample, and DocEngine drafts the accessibility template from it. Available today.

A template is the tagged master document the engine uses to generate PDF/UA output. Until now, someone built it by hand for every new document type: font sizes, column splits, section boundaries, field positions. Slow, precise work. A reviewed draft amortizes across later documents of that type; when the type changes shape, a designer looks again.

Inside the draft flow, an LLM decides what things are: template name, document language, semantic roles, heading hierarchy, which regions are tables. The code keeps every coordinate. If you have read how we built this, you know the boundary. This post is about what to demand before trusting any AI accessibility feature.

Two things to demand

A human review gate. Every draft carries a confidence score per element; anything below 0.60 is flagged. Ask a vendor what happens under their threshold. The right answer names a person who reads the flagged drafts before anything reaches production. “The AI is very accurate” is not an answer.

A validation loop. After DocEngine processes a document, veraPDF validates the result against PDF/UA-1 inside the same workflow, and you see the report: failed checks and fixes, not a green tick. Ask any vendor for that report on your own sample, not a screenshot of a passing run on theirs.

Plenty of AI accessibility features are demos wearing product clothes. Demand those two from any vendor. If the answer is a demo, walk.

One honest limit

Scanned PDFs are not supported yet. The draft flow handles born-digital documents.

Where to see it

DocEngine converts standard PDFs to PDF/UA and now drafts the templates that make it fast. Book a demo and bring your worst PDF. Even the scanned one. We will name what it cannot do before you ask.


Edit on GitHub