Meeting notes are the classic manual chore: someone takes them, someone forgets, and the quality swings with whoever held the pen. I built a system that produces them hands-off after every call: a bot joins the meeting, transcribes who said what, and files speaker-attributed notes into Google Docs before anyone asks.
Note-taking tools mostly ask a human to do something: click record, upload a file, paste a link. The goal here was zero touches: invite one account to a meeting and the notes simply appear afterward. That forces some hard engineering: a bot has to join Google Meet like a real participant, capture clean audio in a headless container, know who was speaking, survive long meetings without running out of memory, and do all of it on infrastructure that costs nothing while no meeting is happening.
A pipeline of small, single-purpose stages: Apps Script watches the calendar and fires a Cloud Run Job per meeting; the job joins, records, and transcribes; Gemini turns the transcript into notes; the notes land in a per-day Google Doc. Everything scales to zero between calls.
A script polls the calendar for meetings the bot account is invited to and launches a Cloud Run Job for each one at the right moment. No servers to keep warm, no cron boxes.
A Playwright-driven headless Chrome joins Google Meet as a real invited participant, handles the join flow, and stays for the duration. One container per meeting, gone when the meeting ends.
A virtual audio sink records the meeting audio cleanly in a headless environment, paired with live-caption scraping so the transcript knows which speaker each line belongs to.
The bot only records while people are actually present, and it archives the audio before transcribing, so a long meeting cannot OOM the container and take the recording down with it.
Google logins do not survive naive automation, so a one-time noVNC flow lets me sign the bot account in interactively and capture the session the headless bot reuses afterward.
Invite one account to a meeting and the work is done: the bot shows up, listens, and files speaker-attributed notes into that day's Google Doc. Because every stage scales to zero, the whole system costs close to nothing when nobody is meeting. Own product, built end to end on Apps Script, Cloud Run Jobs, Playwright, faster-whisper, Gemini, and the Google Docs API.
Book a free call. You will leave with a clear plan for your own AI automation, whether or not we work together.
Book a free call