Automating Digital Accessibility
Building LLM-powered tools that make the web more accessible for people with visual impairments.
A Computing for Good project at Georgia Tech, in partnership with Vision Aid Digital Accessibility Testing.
About the Project
We are partnering with the Vision Aid Digital Accessibility Testing Team to build LLM-powered tools that automate digital accessibility analysis and remediation. Our system analyzes webpage source code, identifies WCAG compliance issues, and generates structured reports with proposed fixes. A secondary tool will use these reports to automate code-level accessibility corrections.
Our Goal
Reduce the time required to produce digital accessibility reports from hours to minutes using large language models, and automate the application of accessibility fixes to front-end code — making the web more accessible for people with visual impairments.
Our Team
Five Georgia Tech students working together to advance digital accessibility.
-
Annika Hildebrandt
Preprocessing & Quality Assurance
Developed the HTML preprocessing approach that classifies accessibility checks as programmatic or LLM-requiring, dramatically reducing token costs. Created and iterated on element-specific LLM prompts across all three WCAG checklists. Evaluated pipeline precision/recall against the manual audit baseline and refined prompts to eliminate duplicate findings.
-
Andrew Yin
Infrastructure & Integration
Built the end-to-end pipeline from raw HTML input to structured accessibility report, integrating programmatic checks with LLM analysis. Handles server deployment on Render and extended the semantic checklists to cover forms and non-text content. Created the web interface for running audits.
-
Cole Niblett
Prompt Engineering & Evaluation
Built the modular LLM prompt pipeline for WCAG accessibility analysis, generating element-specific prompts tailored to preprocessed HTML inputs. Developed a multi-model evaluation harness comparing Claude and OpenAI outputs. Added OpenAI API support to the pipeline.
-
Nicholas Fulton
Web Scraping & Architecture
Implemented the link crawler with configurable depth for multi-page site analysis. Formalized the codebase with abstract base classes and PyPI-compatible package structure for long-term sustainability. Ran initial experiments with open-source models to benchmark against frontier LLMs.
-
Mariana Mendez
Manual Audit & Accessibility
Completed the manual WCAG 2.1 AA accessibility audit of Vision Aid pages, establishing the ground-truth baseline for evaluating automated output. Made HTML/CSS accessibility improvements to the project site. Updated LLM prompts and the report generator to produce specific before/after HTML fix recommendations.
Deliverables
The key outputs our team is building for Vision Aid.
- Accessibility Report Generator — Analyze static webpage code for WCAG compliance issues and produce structured reports with proposed fixes.
- Automated Webpage Editor — Generate code-level fixes for identified accessibility issues, enabling rapid remediation.
- Chrome Browser ExtensionStretch Goal — Apply accessibility fixes directly in-browser for end users.
Lighthouse Scores
Measuring the quality of this website across four key categories.
Performance
Accessibility
Best Practices
SEO
Scores measured via Chrome DevTools Lighthouse on the home page.
Accessibility Audit Tool
Provide an HTML file, or enter a URL to fetch and audit — with optional recursive crawling of nested pages.