← ALL PROJECTS
Cal Hacks 11.0 · Winner, Deepgram Voice Agent Quest

Text to Dot

A handheld device that reads physical text out loud and converts it to real time braille, built for people who are blind or deafblind and can't rely on existing screen-reader style tools.

My RoleHardware + systems integration
Team4 people
Built WithCartesia, Deepgram, Groq, Hyperbolic
Codegithub.com/psycho-baller/text2dot

Overview

Most assistive reading technology is built for online text and doesn't hold up well for everyday situations like reading a menu or a printed page. It also tends to assume the user can rely on audio, which leaves out people who are both deaf and blind. Text to Dot is a three-finger, handheld device with a live camera and a set of solenoids that render Grade 2 braille, so a physical text can be read on the spot in either braille or speech.

What I worked on

  • Hardware control. Wrote the code that controls solenoid actuation, translating decoded text into the correct on and off pattern for each braille cell.
  • Systems integration. Interfaced the hardware over websockets with a vision-language model for transcription and a text-to-speech model for audio output, connecting the camera feed to both the braille output and the spoken output.

How it works

A camera feed captures the physical text, which is passed to an AI pipeline to clean up and interpret. From there, the output is routed either to a text-to-speech model or converted into the standard 6-cell braille pattern and rendered by the solenoids, with a short pause between letters so the user has time to read each one before the device moves on.