A recording of a stand-up becomes minutes with tasks
Audio becomes structure: what was discussed, what was decided, who does what, and what is still open.
What the scenario looks like
- Execute + Data
- LLM
Minutes are written by whoever ran the meeting, which usually means nobody. Two weeks later half the questions are discussed again, because nothing recorded how they ended last time.
What is inside
- A recording — the entry; the file arrives as an attachment.
- Transcription turns speech into text.
- Minutes — the model returns, separately, a summary, the decisions, the tasks and the open questions.
- Formatted minutes assembles a document out of them.
Fields first, prose second. Asking for “nice minutes” straight away gets you a retelling in which decisions and discussion are mixed; splitting into fields forces the model to tell them apart.
The “still open” section is the valuable one
Decisions and tasks are remembered without minutes. What gets lost between meetings is the questions that were left hanging: two weeks later they are asked again. That section is the one part of this scenario not to remove.
What to set up for yourself
- The minutes prompt. A decision currently means something agreed rather than something discussed. That boundary is worth adjusting to how people talk where you work.
- The formatting in the template: headings, section order, what to show for an empty list.
- Where the minutes go. From here they want to be in a knowledge base or a table — then in six months you can ask when exactly the supplier decision was made.
What to connect
Nothing. Transcription and the model are paid for in quarks from the balance. Mind the length: an hour of audio takes minutes to transcribe and then goes into the model whole. A fifteen-minute stand-up is fine; a ninety-minute meeting wants its transcript split.