Note Sensei

Let the phone remember the meeting

Note Sensei started with a familiar moment: a useful meeting ends, everyone agrees to follow up, and then the important details slowly disappear. The idea was to let a phone record the conversation and turn it into notes that were easier to revisit than a raw recording.

The mobile app handled the capture experience while the backend did the heavier work. Audio and session state moved through a realtime path, and an AI-powered service turned the conversation into a more useful summary. The goal was deliberately simple: spend less time reconstructing what was said and more time acting on it.

A small system with several moving parts

React Native and Firebase formed the application layer. The backend used a containerized Node and Express service with Genkit and Google AI integrations, while WebSockets kept the long-running processing flow connected to the client. Docker made the backend easier to package and run as a separate service.

Designing around waiting

The interesting part was not only the generated note. It was the period in between: recording, uploading, processing, waiting, and receiving a result. That flow made progress states and realtime communication feel like part of the product instead of plumbing hidden behind a button.

The lesson

Note Sensei was a compact experiment in giving an AI feature a useful shape. The model could produce text, but the surrounding application had to decide what to capture, when to update the user, and how to make the result feel connected to the meeting it came from.