Sightline

Under the hood, Sightline pairs a deterministic audit engine with LLM reasoning: an on-device analyzer parses HTML with zero network needed and runs a battery of WCA G checks — missing alt text, unlabeled inputs, unnamed buttons, contrast ratios, viewport, headings and document language — producing a 0–100 score, severity-ranked find ings, and a marked-up preview where numbered badges map 1:1 to finding cards. The AI layer calls Meta's Model API (muse-spark-1.3-contributor) for a plain-words verdict with the top 3 fixes first, plus per-issue minimal corrected HTML, with keys validated against the live API, stored encrypted per user, and never shipped or logged. The deterministic core is covered by 36 unit tests, the UI is fully bilingual (EN/ES), and everything was built solo during the hackathon week and released open source under GPL-3.0.