UAB Case Study: Website Accessibility Analyzer
Short Description
Web accessibility remains a persistent barrier for blind and partially blind users, and manual WCAG (Web Content Accessibility Guidelines) audits can be slow, inconsistent, and often skipped under development time pressure. This use case uses ALFIE's AutoML platform and a LLM model to build a Smart Website Accessibility Analyzer that automatically scans URL and flags WCAG violations, such as missing alt text, absent ARIA roles, poor keyboard navigability, and weak colour contrast, returning structured feedback developers can act on.Additionally, it evaluates text readability against standardized metrics to ensure comprehension.
The goal is to make accessibility testing scalable and embeddable in everyday workflows, so issues are caught early rather than fixed retroactively. The tool will use a LLM model configured and evaluated in collaboration with accessibility specialists to analyse web content and provide accessibility feedback. The demonstration will progress from prototype integration and pilot testing, to a full-scale evaluation against defined KPIs covering usability, adoption, and compliance.
Full Description
This use case applies the ALFIE AutoML platform and a LLM model to the automated analysis of web content for compliance with the Web Content Accessibility Guidelines (WCAG), with a particular focus on improving outcomes for blind and visually impaired users. The system ingests URLs and runs automated checks across multiple WCAG criteria, producing accessibility scores and identifying specific violations. The primary objective is to streamline accessibility evaluation by automating the detection of issues that are commonly missed or too time-consuming to catch manually, including missing alternative text for images, absent ARIA (Accessible Rich Internet Applications) roles, weak keyboard navigability, and inadequate colour contrast that may prevent or unnecessarily complicate access for users of screen readers, keyboard navigation, or other assistive technologies. Beyond flagging problems, the tool is designed to guide developers toward more accessible design choices, helping embed accessibility into standard development practice rather than treating it as an afterthought.
This need arises because development teams are frequently under time pressure and may lack specialised accessibility expertise. An automated scanning tool that returns clear, structured, and technically precise feedback lets teams focus effort where it matters most, and integrating that feedback into existing workflows supports earlier detection and cheaper remediation. The system should also help teams prioritise fixes by severity and user impact, enabling incremental, meaningful improvement. This tool is designed as a preliminary automated check, to be complemented by subsequent human evaluation. Several challenges shape the implementation. Regulatory compliance is an ongoing concern spanning WCAG and GDPR, particularly given that the system processes data related to vulnerable user groups, and the platform must remain flexible enough to adapt to evolving AI governance requirements without sacrificing reliability. Model interpretability is also critical: the LLM model's analysis, prioritisation and scoring decisions should be explainable to developers, end users, auditors, and regulators alike, satisfying both technical and business transparency requirements.
The demonstration unfolds in three phases. Phase 1 (content preparation, prompt/workflow development and model evaluation) involves collecting and preprocessing representative sample web content, defining typical accessibility violations in collaboration with accessibility specialists, and developing and testing the instructions, prompts and evaluation criteria used to guide the LLM model in identifying issues such as missing alt text, poor contrast, structural inconsistencies, and missing ARIA attributes.
Phase 2 (prototype integration and testing) embeds the LLM-based analysis into a user-facing prototype where users upload content and receive annotated accessibility evaluations.
Phase 3 (full-scale demonstration and evaluation) simulates realistic development and audit workflows, using varied and complex web content to test adaptability, and evaluates the system.
In conclusion, this use case addresses a persistent and consequential gap in web development: the difficulty of consistently catching accessibility barriers before they reach end users. It does so by using the ALFIE platform and a LLM model to support automated WCAG analysis and provide interpretable, actionable, and regulation-aware feedback. Through a phased demonstration, the use case aims to make accessible design a practical, embedded part of the development lifecycle rather than a late-stage correction. The tool is intended to support, rather than replace, expert accessibility evaluation and user testing, and an automated pass should not be interpreted as evidence of full WCAG compliance, which will later be verified by a human expert.
6 Questions
Asked: 1 month, 1 week ago
By: Uab_Etd(Citizen)
58 Views
Asked: 1 month, 1 week ago
By: Uab_Etd(Citizen)
67 Views
Asked: 1 month, 1 week ago
By: Uab_Etd(Citizen)
69 Views
Asked: 1 month, 1 week ago
By: Uab_Etd(Citizen)
88 Views
Asked: 1 month, 1 week ago
By: Uab_Etd(Citizen)
87 Views
Asked: 1 month, 1 week ago
By: Uab_Etd(Citizen)
95 Views