AdaptEd

The problem

800M+ people have disabilities, yet ~90% of educational videos lack basic accessibility — captions, audio description, sufficient contrast. Fixing it manually per-video doesn’t scale.

How it works

Upload any educational video; AdaptEd analyzes it for gaps, scores it 0–100 against WCAG 2.1, one-click auto-remediates, and produces a downloadable PDF compliance report with a before/after delta. A multi-model pipeline splits the work: Gemini 2.5 Flash for native video understanding, a Llama 3.3 70B agent for transcript-level WCAG reasoning, and Whisper for speech-to-text.

Key features

• 0–100 accessibility score with findings mapped to specific WCAG 2.1 criteria
• One-click remediation: AI captions, contrast enhancement, audio normalization to −16 LUFS
• Braille translation (liblouis UEB Grade 2) rendered inline as a video side panel
• PDF compliance report with WCAG checklist and before/after score delta
• A WCAG-compliant UI that practices what it preaches (15.4:1 AAA text contrast)

Architecture

A React + TypeScript frontend talks to a FastAPI backend, both containerized via Docker Compose. Backend modules split cleanly per model: Gemini analyzer, Gradient (Llama) agent, Whisper transcriber, Braille translator, and an FFmpeg/OpenCV video enhancer.