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