https://github.com/LiamHoweth/HACKATHON?tab=readme-ov-file
Inspiration
People with ALS and other conditions often can’t speak or use their hands. Most communication tools spell letter by letter, and the faster ones need eye-gaze hardware that costs thousands. By the time a reply is ready, the conversation has moved on.
What it does Listens: the partner just talks. Suggests: the two most likely replies appear, plus “Other ways to answer”. Lets you choose with your face: raise your eyebrows to move, drop your jaw to go back, blink long to choose. Speaks: the reply is said in a natural voice.
It’s calibrated to each person’s face, and nothing is spoken unless they choose it. It only needs an iPad.
How we built it Stack: Next.js and TypeScript. Face tracking: MediaPipe, on the device. AI and voice: OpenAI for transcription and replies, ElevenLabs for the voice. Backend and app: a Railway backend, and a native iPad app. Tests: 586 unit tests. AI coding: built with Claude Code and Codex. Challenges we ran into Telling deliberate blinks from natural ones. iPad audio blocking the microphone. Partners interrupting mid-choice. Follow-ups now merge into the open question. Latency: the median fell from 5.1 s to 3.8 s. Accomplishments that we’re proud of Hands-free conversation working on a stock iPad. One blink speaks a whole reply. What we learned
The hard part isn’t generating replies. It’s making sure the person stays in control.
What’s next for Still Here Test with people with ALS and speech therapists. Support phones, and work offline. Personal voice banking.
Log in or sign up for Devpost to join the conversation.