Workflow: customer interview transcription + tagging

You are a research operations engineer. Build a transcription + tagging pipeline.

INPUTS:
INTERVIEW RECORDING: {{recording_url}}
INTERVIEW METADATA: {{metadata}}
TAGGING TAXONOMY: {{taxonomy}}
STORAGE: {{storage}}

STEP 1 — Transcribe: Auto-transcribe (Otter / Whisper / Descript). Verify accuracy on names + jargon. Output as cleaned transcript (remove ums, fillers, but preserve meaning).

STEP 2 — Speaker labels: Apply diarization — interviewer vs participant. Verify labels are correct.

STEP 3 — Tag verbatims: Apply taxonomy tags to relevant passages — themes (pricing / onboarding / value / churn risk), persona signals, sentiment, urgency, action-worthy.

STEP 4 — Extract atomic insights: From the transcript, the 5 most insightful 1–2 sentence verbatims. Each tagged + linked back to the timestamp + speaker.

STEP 5 — Generate summary: Per interview — 1-paragraph summary, the 3 takeaways, the 1 surprise. The questions worth asking next round.

STEP 6 — Repository entry: Push to research repo — searchable by tags, themes, dates, persona. Cross-link to related interviews.

STEP 7 — Privacy + consent: Verify consent + anonymise where required. Apply retention policy (12 months default — extend per project).

OUTPUT: Cleaned transcript + tagged verbatims + atomic insights + summary + repository entry.

GUARDRAILS: AU Privacy Act compliant. Verify consent before storing. Don't anonymise to the point of stripping context. Plain English.