Mantis – AI-Powered Visual QA for Web

Built an AI-powered web accessibility and UI bug detection tool that crawls websites to discover issues before they reach production. Created at Hack the North 2025, Mantis addresses the limitations of traditional testing tools that rely on static rules or DOM inspection and miss visual bugs that only appear under specific interactions or screen states. The system features a BFS-powered crawler with smart link discovery and session management, using Playwright to simulate real user behavior including clicks, scrolls, and viewport resizing. The AI Inspector uses multimodal vision models to detect issues that traditional tools miss: off-screen elements, overlapping components, broken images, and disabled interactions. Key technical achievements include prompt-engineering multimodal models for UI issue detection, efficient async crawling with dynamic viewport capturing, and developer-first experience design with actionable reports that link issues directly to DOM nodes with fix suggestions. The tool integrates seamlessly into CI/CD pipelines and GitHub Actions, enabling developers to catch bugs before merge and reduce QA load through automated visual health checks.