Semantic Headings
Real <h1>–<h6> hierarchy. Never skips levels. One h1 per document. Correct nesting every time.
PDFs are notoriously hard to make accessible. HTML — done right — is natively compliant. SentraCheck converts your documents in seconds with semantic structure, alt text, keyboard navigation, and correct reading order included.
Four stages. Five to thirty seconds end-to-end. Every document passes through the same pipeline — parse, structure, annotate, emit. You get the original PDF unchanged plus a fully WCAG 2.1 AA HTML version, ready to embed.
SentraCheck reads the PDF's positional text and layout data, then rebuilds it from scratch in semantic HTML5 — not converted, not scraped. Nine checks across seven distinct WCAG criteria, handled per document, every time.
Real <h1>–<h6> hierarchy. Never skips levels. One h1 per document. Correct nesting every time.
Every figure, chart, and diagram gets AI-generated descriptive alt text. Decorative images get alt="".
<th scope="col|row"> on every header cell. <caption> describes what the table contains.
Skip-to-content link. Logical DOM order. All content reachable by Tab key without a mouse.
<html lang="en"> declared so screen readers use the correct pronunciation engine.
Spanish, Hmong, Punjabi, and other non-English passages within a document are individually tagged, so a screen reader switches pronunciation correctly mid-document — not stuck reading Hmong with English phonetics.
Minimum 4.5:1 contrast for body text, 3:1 for large text — WCAG AA enforced in CSS.
<main role="main"> wraps content. ARIA landmarks let screen reader users jump between sections.
No "click here" or "read more." Links are meaningful in context and when read in isolation.
Everything above is baked into the output HTML itself, so it works wherever you host it. Link to the optional, free SentraCheck Viewer instead of self-hosting, and readers also get a full accessible reading experience on top — no extra setup on your end.
Text-to-speech with the current sentence highlighted as it's read, adjustable playback speed, and automatic voice switching for inline foreign-language passages.
One-click switch to Atkinson Hyperlegible with wider letter, word, and line spacing — built for readers with dyslexia or low vision.
A highlight band follows the reader through dense text, helping anyone who loses their place — common with ADHD, dyslexia, or low vision.
One tap hides the table-of-contents sidebar for distraction-free reading, without losing access to the rest of the toolbar.
Default, Dark, Sepia, or High Contrast — light sensitivity and low-vision needs vary person to person, so it's not just one on/off toggle.
Search the document by speaking instead of typing — helpful for motor-impaired readers, and a natural fit on mobile.
The table of contents becomes a slide-in panel on phones, with larger touch targets throughout — most residents read agendas on a phone, not a desktop.
Font size, theme, and reading-aid choices persist across every document a reader opens — set them once, not every time.
Two built-in mechanics — Bring Your Own Key and agency-wide SHA-256 caching — mean only truly new, unique PDFs ever cost an AI call. Everything else is instant and free.
By default, SentraCheck handles the Claude API and bundles cost
into your subscription. Plug in your own sk-ant-…
key and AI charges go directly to your Anthropic account at
published rates — zero SentraCheck markup.
Every PDF is fingerprinted on upload. If that exact byte sequence has been converted before — by you, or anyone at your agency — we serve the cached HTML in under a second with no AI call and no cost.
A scanned council packet isn't a PDF with a problem — it's an image with no text at all. Most tools describe the page in one long alt-text caption. We give it a real text layer, built and verified at production scale.
Searchable, selectable, and navigable word by word — not a page-long alt-text description of a scanned image.
Every remediated scanned page is independently verified against ISO 14289-1 by veraPDF — not our own scoring.
Word-level positioning derived from OCR bounding boxes, so reading order and structure survive — not a flat text dump.
OCR misreads happen. Low-confidence pages are flagged for human review — never published silently (~8% of pages, in production).
Tested to 590 pages end-to-end, sustaining roughly 2,400 pages/hour. Interrupted jobs resume instead of restarting from page one.
Pages that already have a text layer skip OCR entirely — only genuinely scanned pages get the OCR pass.
In a live demo, we'll convert a public-records PDF, council agenda, or scanned report from your own agency. WCAG 2.1 AA HTML with semantic structure, alt text, and full keyboard navigation.