Accessible documents: from PDF to the web and back, with no manual layout
An agent learns how each kind of document is built and turns it into an accessible, editable web page. The PDF is then generated back from it, signed.
The problem
“It has to go on the website and as a PDF, and it has to be accessible.” Behind that sentence someone is laying out the same document twice, page by page. And the PDF still cannot be read by a screen reader: no headings, no reading order and tables that are just drawings.
It happens in public administrations and in any organisation that publishes long documents: reports, annual reviews, guides, regulations. Every change to the text means touching both versions, and they end up saying different things.
How it is done today
Someone opens the PDF, copies the text to the website piece by piece, rebuilds the tables, puts the notes and links back and marks up the headings. With every new document, again. When the content changes, it gets fixed in one place and forgotten in the other.
In our own company two people were dedicated to laying out documents by hand.
How we solve it
- You give it a document of a given kind: a report, an annual review, a guide.
- The agent studies how that kind of document is built: heading levels, tables, notes, figures, lists and what repeats on every page.
- With that structure it turns it into native HTML: headings that are headings, tables that are tables and text that can be searched, read by a screen reader and edited.
- A person reviews the content on the web and corrects whatever is needed, now on a single version.
- The PDF is generated back from the HTML, with a digital signature.
- The next document of the same kind no longer has to be studied: it is converted with what was learned.
Result
2 → 0people doing layout by hand
In our own company two people used to lay out documents by hand. Today documents lay themselves out, and a person validates the content and signs.
What it takes
Timeline
It depends on how many kinds of document you have and how regular they are. A kind with a stable structure is learned quickly; one that changes from document to document needs more review. The exact timeline comes out of the assessment, with a sample of your documents.
Data and access
A sample of each kind of document, ideally with its original source if it exists. Permission to create content on your website or portal. And your visual identity, for the PDF generated back.
Your tools
Your content management system or portal, where the HTML is published, and the certificate of whoever signs the PDF. The source documents can come from any software.
How it is built
It is not OCR. The agent infers the structure of each kind of document and generates semantic HTML: headings in order, tables with their headers, lists and notes.
The HTML is the single source. The PDF is generated from it and signed with the person's certificate, which never leaves their computer.
The resulting page can be analysed like any other with our automated accessibility audit: a headless browser and axe-core, against the WCAG 2.2 level A and AA criteria. Automated checks only detect part of the problems, which is why the report always says what has to be reviewed by hand.
Frequently asked questions
Is it OCR?
No. OCR pulls letters out of an image. Here the agent learns the structure of each kind of document and generates a real web page, with its headings, tables and notes.
Does the result meet accessibility standards?
The HTML is born with an accessible structure and can be analysed with automated tools, which only detect part of the problems. The rest is reviewed by a person. We do not promise an automatic accessibility statement.
What happens if the document changes?
It is corrected on the page, which is the single source, and the PDF is generated and signed again. There are no longer two versions to maintain.
Does it work for any document?
It works best with kinds of document that repeat: periodic reports, annual reviews, guides. A one-off, very irregular document can be converted, but needs more review.
What does Julditec do?
We implement the agent, adapt it to your company, train your team and maintain it. The day-to-day work is done by the agent; deciding and approving stays with your team.
Related automations
Signed documents
Reports and documents as PDFs with your branding and a digital signature, with nothing printed or filled in by hand.
Public tenders
The tenders that fit your company, matched against your CRM, with the outreach to public bodies prepared.
Part of Legal and compliance