Build a meeting-action extractor with source references

A fluent action list can assign work nobody accepted. Every proposed task should point to the transcript evidence and make missing owner or deadline visible.

Reject an unsupported action

The automation lab uses two authored transcript lines and proposals.

from automation_cases import meeting_action_case

result = meeting_action_case()
assert result["accepted_indices"] == [0]
assert result["review_indices"] == [1]
assert result["actions"][0]["source_line"] == 1
assert result["transcriber_or_model_executed"] is False

The draft action has owner and due phrase. “Approve budget” lacks those facts and routes to review. No audio or model ran.

Build an evidence schema

Store meeting/source ID, timestamp or line, speaker, action verb/object, owner as spoken, due text, dependencies and confidence. Preserve a short source excerpt under the meeting’s access policy. Separate decisions, questions and suggestions from commitments.

Validate participant identity against the meeting roster without guessing from first names. Treat relative dates using the meeting date and timezone, then show the resolved date for review. Require a human confirmation before creating tasks or sending reminders.

Evaluate action precision/recall, owner and due accuracy, unsupported proposals and review workload across accents, overlapping speech and vague language. Protect recordings and transcripts with consent, retention and role access.

The Generative & Agentic AI course links multimodal evaluation to approval-first automation.

Exercise

Write a ten-line synthetic meeting with actions, decisions and ambiguity. Extract structured proposals, cite every line and prove uncertain owners never create external tasks.

Continue learning

This article is part of the Business automation with AI sequence. Use the neighbouring tasks when you need the prerequisite or the next application.

• Review the prerequisite or neighbouring task in Monitor a no-code automation for silent failures.
• Continue with Automate a knowledge-base update proposal with review.

Reference: NIST AI RMF Generative AI Profile.