Using Grok Bot to pull tasks and owners from meeting recordings
The author outlines a pipeline of faster-whisper, pyannote and Instructor feeding Grok Bot to extract timestamped decisions and tasks, with manual verification.
ONE HOUR ON A CALL. THREE DIFFERENT VERSIONS OF WHAT WE AGREED ON. I’D GIVE GROK BOT THE RECORDING AND THESE 3 GITHUB REPOS. after a meeting, everyone remembers a different deadline and nobody owns the task. here’s the workflow i’d build around grok bot: 1/ faster-whisper · 26.1K⭐ turn speech into timestamped text so i can jump back to the exact moment something was said. → https://github.com/SYSTRAN/faster-whisper 2/ pyannote. audio · 10.7K⭐ separate the recording by speaker. then match speaker labels to names so a commitment doesn’t get assigned to the wrong person. → https://github.com/pyannote/pyannote-audio 3/ Instructor · 14.1K⭐ get structured output from a connected model: task, owner, deadline and supporting excerpt. validating the format doesn’t prove the content is correct, so references to the recording still matter. → https://github.com/567-labs/instructor i’d connect the tools through scripts and a compatible api, then give grok bot one transcript: extract decisions, tasks and open questions with timestamps. don’t invent owners or deadlines. i’d check the recording before sharing the list. did we agree, or just discuss it?