Side Projects

Real-time Japanese speech transcription with English translation, summaries, and response suggestions. Built for foreign residents in Japan who need to follow complex conversations.

I live in a rural farming neighborhood in Japan. While I feel comfortable using Japanese in most of daily life, neighborhood meetings were always a completely different story. I'd walk out feeling like I understood maybe 5% of what was discussed. The vocabulary, the speed, the local dialect. It was overwhelming and frustrating. I built Live Kaiwa to solve that problem for myself, and for anyone else living in Japan who has experienced that same feeling.

How it works:

1. Audio capture — The browser microphone streams audio via WebRTC to OpenAI's Realtime API, or records 3-second chunks for local Whisper transcription.
2. Transcription — Japanese speech is transcribed in real time. A multi-pass system re-transcribes buffered audio at higher accuracy and replaces earlier results.
3. AI analysis — Each batch of new transcript lines is sent to an LLM which returns translations, summary bullets, and suggested responses, all streamed back.
4. Local persistence — All session data stays in the browser. Nothing is stored on any server.