meeting-decision-log-extractor

When to invoke

• You have a meeting transcript (or notes) and need structured decisions/action items.
• You want to standardize follow-ups and ownership.

Inputs needed

• Transcript text file (UTF-8).
• Optional participant list.

Workflow

1. Segment transcript into lines and identify speaker prefixes.
2. Detect decision statements (e.g., “we will…”, “decided…”, “approved…”).
3. Detect action items (e.g., “ACTION:”, “owner”, “by <date>”, “TODO”).
4. Extract owner and due date with lightweight heuristics.
5. Output Markdown with two tables: decisions and actions.

Output format

• Decisions table (id, decision, owner, date)
• Action items table (id, action, owner, due date)
• “Uncertain extractions” section

Guardrails

• Mark low-confidence rows as uncertain; do not invent owners/dates.
• Keep raw quoted text snippets so a human can verify.

Reference code

• extract_decisions.py

  reads a transcript (.txt) and writes Markdown.