Accessible Documents Initiative

Early access, in pilot

AltBridge

Every image described. One upload, one shared link.

A subject expert review tool for all images in STEM documents. Upload the whole PDF, no more cropping images one at a time. AltBridge extracts every figure with section context, drafts alt text per figure, and produces one shareable review URL. Faculty review in the browser; output is a Word file with embedded figures and accepted alt text.

Try AltBridge with your own PDF How it works

At a glance

  • What it is. A web page that lists the images from a document and asks an expert to confirm or correct the alt text for each one.
  • Who it is for. Accessibility remediators, faculty, librarians, publishers, and clients who review alt text decisions.
  • What you need. A modern browser. The demo runs on this device only. Nothing is sent to our servers.
  • How long it takes. The demo has six images and takes about five minutes to walk through.

Try the demo

The demo is a small set of figures that shows how the tool feels to use. You can stop at any point. You can come back later. Your responses stay in your browser on this device.

Open the demo

What you will see in the demo

The demo has six sample images. Each one shows a different review situation:

  1. An image where the alt text is a clean match.
  2. An image where the alt text is clearly wrong.
  3. An image where the AI flagged a problem that is not really there.
  4. An image where it is not clear, and an expert needs to decide.
  5. An image with no alt text at all.
  6. An image where colour carries meaning that the alt text does not capture.

You confirm or correct each one. You can add a short note if guidance is needed.

What happens to your responses

  • Responses are saved in your browser on this device only.
  • Nothing is sent to Accessible Documents Initiative servers.
  • You can reset the demo at any time by clearing browser storage for this site.
  • You can share the demo URL with colleagues so they can try it on their own device.

How it works

The tool follows a simple three-step flow. Each step has one job. You only see what is needed at that step.

  1. Step 1. The remediator prepares a session

    The remediator builds a list of figures from one or more PDFs. Each entry has the current alt text, an AI suggestion, and the reason for the suggestion. The tool creates a unique link for the session.

  2. Step 2. The expert opens the link

    The expert opens the link in a browser. There is no install and no login. They see each image with its current alt text and the AI verdict. They confirm, override, or mark for a second look. They can add a short note for any image.

  3. Step 3. The remediator exports the responses

    When the expert is done, the remediator downloads a CSV file. It has one row per image with: page number, figure number, current alt text, AI suggestion, expert verdict, and expert note. The remediator loads this into their remediation tool and applies the corrections.

Who this helps

  • Remediators working at volume

    A textbook chapter can have 40 images. A full batch can have several thousand. The tool lists every image on one page so the reviewer never has to open the source PDF to compare.

  • Remediators working with subject experts

    A biology diagram, a physics problem, or a chemistry workflow often needs a faculty member or course author to look at it. This tool replaces the email thread with one shared link.

  • Client and publisher reviewers

    A client reviewing an accessible document can open the link, check the alt text decisions, and return feedback as a CSV.

Built to be accessible to the people who use it

Many accessibility tools are not themselves accessible. This one is built so that a blind or low-vision remediator can use it end to end with a screen reader.

  • Every verdict (match, ambiguous, mismatch) is announced as text. Colour is supporting, never the only cue.
  • Every form field has a visible label.
  • Each figure is its own heading level so screen reader users can jump from image to image.
  • The AI reasoning is written as full sentences in reading order. It is not hidden in a tooltip or popover.
  • All actions work from the keyboard. There is no mouse-only zone drawing.

This is a quality bar, not a headline feature. Remediation tools should be usable by blind remediators, not only usable for producing accessible output.

Privacy and security

  • The session URL is a private share link. Anyone with the URL can open the session. Treat the link like a password.
  • In the demo, all responses stay in your browser on this device. Nothing is sent to Accessible Documents Initiative servers.
  • The next phase will add expert tokens with expiry and server-side storage. When that ships, existing session URLs will continue to work.
  • Do not post session URLs in public channels.

Status and pilots

AltBridge is in early access. It was first used in production on a 189 PDF Slizvon accessibility QC batch in April 2026. In that batch, one reviewer worked through 3,926 figures in about two hours, and surfaced 69 corrections that the AI had under-weighted.

The packaged version is open to further pilot use by remediation teams, university disability-services offices, and publishers in the Global South and elsewhere.

To discuss a pilot or request a real session for your batch, email deepa@equitabledocs.org.