Accessibility Checker for Blind Visitors of Websites

 
  • UAB use case
  • Interacts with automl using a screenreader + Speech to text

Input

  • website url or uploads html/css files

Requirements

  • missing alternative text for images
  • lack of proper semantic HTML structure
  • insufficient colour contrast
  • absence of ARIA roles.
  • Ensure that the website is navigable and understandable using popular screen readers.
  • compliance with WCAG 2.1 (Web Content Accessibility Guidelines) criteria relevant for blind users.
  • Simulate screen reader behavior

Output

  • Provide specific recommendations to fix detected issues.
  • Dashboard with an accessibility score, highlights etc
  • List of issues by severity
  • WCAG 2.1 compliance

Clarifications

  • Do we need the automl system to generate a model or just use the automl system to do this. Eg: we would run an llm over the users code and then have results.
    • There is no normal “model” that can do this reliably.
    • We need two kinds of models essentially - one thats an LLM like chatgpt that can do many things and another that does a model search + refinement over existing architectures to find the best model.
  • How would user intera ction using a screen reader look like? For the output that is

Meeting Notes

  • 2025-05-09
    • UAB Barcelona
    • People
      • Kate - communication
      • Chiara - anthropologist, vulnerable groups
      • Hari S - Philosophy of science
    • Difficulties
      • voice interaction
      • cannot type
    • Person who will use automl system - designer user
    • Ethical models and suggestions