Grounded VLM design reviewer that screenshots UI and validates findings
It runs headless Chromium to capture deterministic screenshots and DOM geometry, then queries a vision‑language model to critique the UI against your design system. Findings that reference routes or elements not captured are automatically dropped, ensuring grounded, actionable feedback. The repo ships a Rust crate for perceptual duplicate detection and calibration tooling, making it CI‑ready. It’s aimed at teams that want reliable, AI‑driven design reviews.
View on GitHub →apatureai/judgment-engine