Meeting page with timestamped notes and a transcript.
Customer
People whose day is calls and conversations they need to remember (product managers, founders, consultants, researchers) and who lose the details because note-taking competes with listening.
Problem
Recording is easy. Everything after it is not. Export the audio, run a transcription app, clean it up, file it somewhere you'll find it. Each step is small, but together they mean recordings pile up untranscribed and the details people recorded them for are gone.
Outcome & measures
Press Record in a browser tab, watch the transcript appear as people speak, ask a question about what was just said, and press Stop. A clean transcript lands in your notes with nothing else to do.
Zero-touch after Stop, from recording to a filed transcript
Accuracy at least as good as the desktop app it replaced
Live text within about 3 seconds of a pause (1.4 to 1.9 seconds in the first test)
No audio or transcript lost to a crash or a closed tab
What's next · Dec 2026
Better accuracy on names and uncommon terms, using the vocabulary already in my notes.
Why I built it
Recording was easy. Everything after it was the problem. I wanted a tool where pressing Stop was the last thing I did.
How it came together
Before any app code, an agent ran a test on real audio: could a local speech model match the desktop app I used, fast enough to feel live? It passed, so the build went ahead. Eight plans followed, from the first recorder to highlights on the live transcript.
What I learned
Prove the hard part first. That test settled the risky question on day one, and everything after it was ordinary building. Writing “zero-touch after Stop” as the goal also kept half-automated features from sneaking in.
