Transcript Intelligence

Conversation Detail

workflow planning discussion

Discussing how to analyse transcription data

Behnam ยท 2026-07-01 16:34:02 to 2026-07-01 16:34:58

Review: Pending Review

Summary

Behnam explains that with working but imperfect transcription, he is thinking about the best way to analyse it. He finds conversation extraction on three-hour chunks messy and not useful, with irrelevant items picked up and serious conversations diluted. He worries about trying to process full context in one go or even half a day at a time. Another participant observes that the current approach picks up irrelevant things, and asks what Behnam thinks. They reference that Behnam previously said Gemini was better for long-context summarisation and ask whether he still holds that opinion. The session ends with one participant ending the recording for the day.

Facts

  • Working transcription is currently imperfect.
  • Three-hour chunk conversation extraction is producing messy or irrelevant results.
  • There is concern about running into difficulty with full-day or half-day context.
  • Gemini was previously recommended by Behnam for long-context summarisation.

Tasks

  • Decide on a better approach for analysing long recordings.
  • Re-evaluate whether Gemini is still the best tool for long-context summarisation.

Issues

  • Current extraction picks up irrelevant things for the sake of it.
  • Deep/serious conversations are getting washed out.
  • Full-context processing may hit technical limits.