AI-Powered Web Accessibility Analysis Tool for RGAA Compliance

La Mêlée's Hackathon - BPCE Solutions Informatiques Project - 1st Place (done in my personal time) Objective: Automate the testing of RGAA criteria to improve web accessibility for users, including those with disabilities, while saving time for development teams. Challenge: Only ~30% of RGAA criteria can be automatically tested with current tools. Manual reviews are time-consuming, and AI could help expand coverage while using resources responsibly. Approach: • Analyzed RGAA criteria to determine which rules required AI intervention. • Selected three technical thematics for automation: images, links, and website structure. • Developed two frontend prototypes with identical visuals but with one respecting the criteria and one not, to illustrate differences. • Built a web application where users can input any website link and get an automated analysis for the three selected criteria: • Images: AI generates descriptions, then an LLM evaluates compliance with RGAA rules. • Links: Context and link text are analyzed by an LLM to determine conformity. • Website Structure: Multimodal LLM assesses both visual layout and code structure. • Added a cost estimation page for each request (financial and environmental impact). • Displayed detailed element-level results, highlighting non-compliance and providing explanations. • Implemented the solution using Azure OpenAI for model execution. 💡 Skills Gained: • Deep understanding of digital accessibility and the challenges faced by people with disabilities • Knowledge of RGAA coding standards and compliance requirements • Ability to analyze which problems require AI, prioritize criteria, and select technical approaches • Hands-on experience with Azure OpenAI and generative AI for practical applications • Enhanced data analysis, problem-solving, and solution design skills in a real-world, responsible AI context