Coaching-call recaps, 90% automated

Problem

After every coaching call, coordinators wrote a recap for the coach and each attendee by hand, plus a quality review and attendance. On heavy days that was three hours of batching, and recaps went out late.

What I built

• A pipeline that picks up each finished recording, pulls the transcript, drafts the recap and a coaching-quality report with a language model, posts a review card to the team, and sends the approved recap to attendees in the community platform.
• Attendance written to the performance database from the same run, with a nightly self-check that repairs any call the pipeline missed.
• Approval kept human: a coordinator reviews and approves in one click. The automation drafts, the person decides.
• Fixed the subtle bugs that show up at scale: two sessions in one meeting room being scored against the wrong recording, and a hung sub-run that silently skipped attendance.

Result

Recap creation is about 90% automated with a human in the loop for the rest. Coordinator batching time dropped from about 3 hours to about 1 hour on heavy days, and recaps land the same day.

What it shows

Automating the drafting and keeping the judgment: the design that people actually trust.