AccessFix AI

AccessFix AI — Don't just fix accessibility issues. Prove users can complete their tasks. AI-generated accessibility fixes can appear successful while silently introducing new barriers. AccessFix AI addresses this problem by combining Meta Muse Spark's code generation with independent, evidence-based browser verification. The system identifies blocked keyboard interactions, generates bounded source-code patches using Meta Muse Spark, and evaluates the results through Playwright task testing and axe-core accessibility audits. A deterministic acceptance gate accepts improvements only when measured requirements pass and rejects changes that introduce regressions. In our recorded experiments, Project Tabs improved keyboard task completion from 0/2 to 2/2 while preserving pointer functionality. Another repair, Notification Switch, achieved the same keyboard improvement but introduced a 1.19:1 text contrast violation. AccessFix AI detected the regression, rejected the patch, and preserved the original implementation. Our public demo provides an interactive, read-only exploration of actual recorded experiments, including source patches, screenshots, browser task results, and final decisions. The goal is to move AI-assisted development beyond generating plausible code toward verifiable, safer software changes.