HearEase AI

Inspiration

Many people with hearing difficulties can follow that information exists — a class announcement, an appointment letter, a government form — yet still struggle to understand it and act on it. Fast speech, dense wording, and buried deadlines create a real accessibility gap. We built HearEase AI for GatewayHacks 2026 (Accessibility & Health) to close it: an AI-powered clarity layer between complex communication and the user.

What it does

Paste any difficult text and HearEase transforms it into four layers — WHAT HAPPENED? (summary), WHAT MATTERS? (dates, times, warnings, contacts as visual cards), WHAT DO I NEED TO DO? (action items), and WHAT SHOULD I ASK? (copy-ready follow-up questions). Every section has "Explain this more simply" with Simple / Very Simple / Step-by-Step levels, plus an accessibility settings panel (large text, high contrast, reduced motion), saved analyses, and honest demo-session metrics. Three fictional demo scenarios (classroom, clinic, public service) make the live demo one click.

How we built it

React + TypeScript + Vite + Tailwind frontend with React Router; FastAPI + Pydantic backend exposing a strict structured-JSON contract (summary, key_points, important_details, action_items, questions, clarity…) so the UI never parses free prose. An AIProvider abstraction lets the app run a deterministic offline DemoProvider (no API key, demo-proof) or swap in any OpenAI-compatible model via server-side key. No secrets in the repo, privacy-first (nothing saved unless the user presses Save), deployed on Vercel.

Challenges

Reliability under live-demo pressure: we made Demo Mode fully offline with a frontend fallback, so the demo survives API or network failure. Respectful simplification was a design challenge — clear language without infantilizing the user. We also kept metrics honest: only current-session stats, never fabricated user numbers.

What we learned

Accessibility is about actionability, not just availability, of information. Structured AI outputs beat clever prompts for rendering reliable UI. And judging criteria reward focus: every feature serves the core accessibility problem — nothing "AI" for its own sake.

Built With

• fastapi
• openai
• pydantic
• python
• react
• rest-api
• tailwind-css
• typescript
• vercel
• vite