AAC Board AI

Inspiration

Communication is a basic human need, yet for people who can't rely on speech, even building a short sentence can take immense effort. I wanted to explore how Chrome's built-in AI could make that process faster, more expressive, and more personal — all while keeping data private and on-device.

What it does

AAC Board AI is a communication board for people with speech or language disabilities (including ALS, autism, and cerebral palsy).

It lets users tap pictograms to build messages — like any AAC board — but it goes further, bringing those messages to life with Chrome's on-device AI:

Proofreader API refines telegraphic utterances (e.g., “I want go grandma”) into natural speech (“I want to go to Grandma’s”).
Rewriter API adjusts tone — neutral, formal, casual, or custom (e.g., “playful and polite”).
Translator API makes the same message instantly available in any supported TTS language.

How we built it

AAC Board AI is built with React 19, Vite, and Material UI 7, using IndexedDB for fully local data storage.

At its core is a schema-validated Open Board Format (OBF) importer written in TypeScript with Zod, ensuring compatibility with community-authored AAC boards.

Imported boards are unpacked into an on-device database (aac-board-db) with structured tables for boardsets, boards, and symbol assets.

On top of that, a collection of typed React hooks orchestrates Chrome’s Built-in AI APIs — Proofreader, Rewriter, and Translator — allowing messages to be corrected, re-toned, and translated instantly and privately, all locally.

Challenges we ran into

I experimented with the Prompt API to generate sentence completions based on the current message and available words. It worked about 80 % of the time and was impressive when it did — but the remaining 20 % produced illogical results (“I eat ocean”). For an assistive communication tool, that margin of error is too high.

Accomplishments that we're proud of

The project provides real value to people who can’t speak — reducing effort, increasing speed, and helping express personality. It also helps speech therapists translate and reuse existing boards instantly.
The open-source implementation serves as a transparent learning resource for developers building accessible, privacy-respecting AI tools.

What we learned

Built-in AI enables a new class of applications that are private, fast, and expressive. I'm continuing to explore how these models can enhance assistive communication tools.