GrapeSEED
2025–presentFounding Designer & Engineer, AI Product
I’m building GrapeSEED Tutor, an app that guides young children from early phonics and reading to real spoken conversation in English. As the founding designer and engineer, I own the product across the stack: the interaction design of a child’s first conversation, the real-time voice pipeline and interactive lessons behind it, and the backend and tooling that keep it running.
Across the stack
The app children learn in
The full client application, built from zero: interactive video lessons, phonics, reading and handwriting practice, and real-time AI voice conversation, across more than a dozen activity types from listen-and-write to pronunciation scoring. A voice pipeline, lesson player, and modular activity engine drive it all, shipped through a GitHub Actions and EAS pipeline with over-the-air updates and native builds.
The backend that powers it
A GraphQL service with row-level security and automated migrations, deployed on Vercel for a two-person team. I designed the data model, auth flows, and content-delivery architecture.
The tools that fill it
An internal CMS where the curriculum team authors everything the app teaches: lessons, readers, phonics, and interactive video activities. It has a Mux pipeline for lesson video and AI-generated illustration art. Its centerpiece is a visual conversation designer built on a state-machine model: non-engineers compose the tutor’s dialogue, AI generates the branching transitions, and the flow can be tested live against the voice tutor.
Earlier
I first joined GrapeSEED as a research intern in 2021, to find out whether conversational AI could teach English on its own. The prototype did, roughly, what the product does today: open-ended conversation, structured lessons and pronunciation feedback for young learners. I came back in 2025 to build it properly.
View the 2021 prototype on GitHub ↗