Skip to content

Parse meeting transcripts into typed action-item arrays

Turn messy multi-speaker call notes into arrays of decisions, owners, due dates, and open questions matching your PM tool.

Copy the prompt and paste it into the AI assistant you already use.

Checklist for clean nested array extraction

  1. 01

    Keep array item objects flat

    Use concise per-item keys like task, owner, and due_iso rather than deeply nested sub-objects.

  2. 02

    Normalize dates to ISO 8601

    Pass meeting_date as an input parameter so relative phrases resolve cleanly or stay null.

  3. 03

    Distinguish ideas from commitments

    Define criteria separating firm action items from casual brainstorming remarks.

  4. 04

    Handle empty lists gracefully

    Ensure meetings with zero tasks return an empty array [] rather than placeholder items.

Common questions

How does the endpoint resolve words like next Tuesday?

Pass a call_timestamp input parameter alongside the transcript so relative dates map accurately to ISO-8601 strings.

What if no owner was assigned to a task?

Making the owner field nullable lets the extractor leave unassigned tasks explicitly null for team review.

Can one call extract decisions and action items separately?

Yes. Your top-level JSON object can hold distinct typed arrays for decisions, action_items, and open_questions.