AdaptEd — AI Accessibility Agent for Educational Videos

Making education accessible for everyone, automatically.

800 million+ people worldwide have disabilities. 90% of educational videos lack basic accessibility features. AdaptEd fixes that.

AdaptEd is an autonomous AI agent that analyzes educational videos for accessibility gaps and automatically remediates them — generating captions, translating to Braille, enhancing visual contrast, normalizing audio, and producing WCAG compliance reports — powered by DigitalOcean Gradient™ AI.

What It Does

1. Upload any educational video (lecture, tutorial, course)
2. AI Analysis — Gemini analyzes the full video for accessibility issues, while the DigitalOcean Gradient AI Agent (Llama 3.3 70B) performs transcript-level WCAG reasoning
3. Accessibility Score — Get a 0-100 score with detailed findings (critical/major/minor) mapped to WCAG 2.1 criteria
4. Auto-Remediate — One-click fixes:
   • AI-generated captions via Whisper speech-to-text
   • Braille translation using liblouis (UEB Grade 2) — rendered as a side panel in the video
   • Visual contrast enhancement
   • Audio normalization to -16 LUFS
5. Compliance Report — Download a PDF accessibility report with WCAG 2.1 checklist and before/after score comparison

Tech Stack

Layer	Technology	Infrastructure
Video Analysis	Gemini 2.5 Flash	Google AI
Transcript WCAG Analysis	Llama 3.3 Instruct (70B)	DigitalOcean Gradient™ AI Agent
Speech-to-Text	OpenAI Whisper	DigitalOcean Droplet
Braille Translation	liblouis (UEB Grade 2)	DigitalOcean Droplet
Video Processing	FFmpeg + OpenCV + Pillow	DigitalOcean Droplet
Backend	FastAPI (Python)	DigitalOcean Droplet
Frontend	React + TypeScript + Tailwind	DigitalOcean Droplet (nginx)
Containerization	Docker	Docker Compose

DigitalOcean Gradient™ AI Usage

AdaptEd uses a multi-model architecture leveraging DigitalOcean Gradient™ AI:

• Gradient AI Agent (Llama 3.3 Instruct 70B): Performs transcript-level WCAG 2.1 accessibility reasoning — analyzes speech patterns, pacing, content descriptions, and navigation structure. Findings are merged with Gemini's video-level analysis for comprehensive results.
• DigitalOcean Droplet (s-2vcpu-4gb): Hosts the FastAPI backend, Whisper inference, FFmpeg video processing, and nginx serving the React frontend.
• Multi-model pipeline: Gemini handles native video understanding (visual content), Gradient Agent handles transcript reasoning (speech/text content), Whisper handles transcription — each model doing what it's best at.

Quick Start

Prerequisites

• Python 3.10+
• Node.js 18+
• FFmpeg installed (brew install ffmpeg or apt install ffmpeg)
• liblouis installed (brew install liblouis or apt install liblouis-bin)
• Gemini API key

Backend

cd backend
cp .env.example .env
# Fill in GEMINI_API_KEY and GRADIENT_AGENT_KEY in .env
pip install -r requirements.txt
uvicorn app.main:app --reload

Frontend

cd frontend
npm install
npm run dev

Docker

docker-compose up

Visit http://localhost:5173

Project Structure

adapted/
├── backend/
│   ├── app/           # FastAPI application
│   │   ├── main.py    # Entry point
│   │   ├── config.py  # Settings
│   │   └── models.py  # Pydantic schemas
│   ├── routers/
│   │   └── video.py   # API endpoints
│   ├── core/
│   │   ├── gemini_analyzer.py      # Gemini video analysis
│   │   ├── gradient_agent.py       # DO Gradient AI Agent integration
│   │   ├── whisper_transcriber.py  # Whisper STT
│   │   ├── braille_translator.py   # liblouis Braille translation
│   │   ├── video_enhancer.py       # FFmpeg + Pillow processing
│   │   └── report_generator.py     # PDF reports
│   ├── Dockerfile
│   └── requirements.txt
├── frontend/
│   ├── src/
│   │   ├── components/  # React UI components
│   │   ├── services/    # API client
│   │   └── App.tsx      # Main app
│   └── package.json
├── docker-compose.yml
└── README.md

Accessibility Standards

AdaptEd evaluates against WCAG 2.1 guidelines:

• 1.2.1 Captions (Prerecorded)
• 1.2.3 Audio Description
• 1.2.5 Audio Description (Prerecorded)
• 1.4.3 Contrast (Minimum)
• 1.4.6 Contrast (Enhanced)
• 1.4.7 Low or No Background Audio
• 2.2.2 Pause, Stop, Hide
• 2.3.1 Three Flashes or Below Threshold

Design System

AdaptEd's UI is built with a WCAG-compliant color system:

• Brand seed: Teal (#0D9488)
• Primary text: 15.4:1 contrast ratio (AAA)
• Secondary text: 7.0:1 contrast ratio (AA)
• Accent text: 11.1:1 contrast ratio (AAA)
• No purple/indigo gradients — an accessibility tool that practices what it preaches

License

MIT License — see LICENSE

---

Built with DigitalOcean Gradient™ AI for the DigitalOcean Gradient™ AI Hackathon 2026.