WordloopWordloop
WorkDeliveredLive CaptureTechnical DesignMilestonesFull Pipeline

ML — Post-Meeting

Slice 4b ML.

Slice 4: Post-Meeting Reprocessing + Speaker ID

Owner: ML engineer Domain: ML Complexity: M Status: ✅ Done

The automatic pipeline that runs after recording stops.

Tasks

publish TranscriptionJobMessage and MeetingTerminatedMessage on session stop

  • consume MeetingSessionTerminated — drain AssemblyAI buffer; send final segments via REST
  • ML: skip task extraction when existing live tasks present
  • Core: POST /transcriptions/{id}/remap-speaker and POST /transcriptions/{id}/identify-speakers
  • ML: speaker centroid computation + identify + remap pipeline

Test cases:

TestLocationAssertion
Post-meeting pipeline triggers automaticallytest_systemRecording stop → transcript segments replaced with higher-accuracy version
Tasks from live session preservedtest_systemAfter post-meeting processing, user-created and system tasks still present
Speaker labels resolved to person_idtest_coreCosine similarity match → segment person_id updated
Talking points promoted to is_final: truetest_systemAfter post-meeting processing, talking points have is_final: true
Headline generatedtest_systemMeeting has non-null headline after post-meeting processing

Core — Post-Meeting

Slice 4a Core.

Full Meeting Experience

Milestone 3 integration overview.

On this page

Slice 4: Post-Meeting Reprocessing + Speaker IDTasks