Echo

The problem

Meetings disappear. Someone takes half a page of notes, the decisions stay in one person's head, and a week later nobody can say what was agreed or who is meant to be doing it.

What I built

We built a recorder you press once: an ESP32-S3 board streams the audio to a backend that separates the speakers, writes the notes, pulls out the action items and makes every past meeting searchable. A dial on the device picks the kind of meeting, and that choice decides which model writes the notes.

What it does

• One Button, One Dial. Press the button to start recording, turn the dial to say what kind of meeting it is, and the little screen tells you it is listening.
• Knows Who Said What. The audio comes back split by speaker, so the notes can say who agreed to what instead of just what was said.
• Searches Every Meeting. Ask what was decided about something and it looks back through everything you have recorded, not only the last call.
• Two Clouds, One Job. Each step goes to whichever provider does it best, Superlinked or Alibaba Cloud Qwen, and either one takes over when the other fails.

Demo

Photos

The full engineering is on GitHub.