Summary
Peter Yang shows how to build an AI language coach that teaches Japanese through live spoken conversation, arguing that voice is on its way to becoming the default way people interact with computers. The finished app is a phone-friendly web app with 10 lessons of 10 travel phrases each, illustrated with diorama-style art, and a character named Yuki who coaches pronunciation and then role-plays a day in Tokyo.
The build follows six steps: explore the idea with AI, draft a spec and design, build the first version, improve the app, improve the lessons, and launch. The video is sponsored by Google and uses Gemini’s Live API for real-time conversation, Nano Banana for scene images, and Google’s Antigravity and Claude Code for development. Early on, he has the AI do research and ask clarifying questions before writing any code.
Practical details include creating a Gemini API key in Google AI Studio and storing it in a local .env file rather than pasting it into a chat for security. In the iteration step, Claude Code uses browser and computer control to test the app, and a microphone problem is traced to the wrong input device. The same approach can be adapted for any language, and the video offers a useful model for solo builders creating personal AI apps.
📺 Source: Peter Yang · Published October 05, 2026
🏷️ Format: Hands On Build
