Productivity · Meeting AI · v2

Meeting to
Action

"Normalize messy transcripts, detect meeting type, extract structured data, track commitments across weeks, and ship a weekly digest — one tool."

View on GitHub API Docs

$ python main.py batch \
  --dir ./transcripts/ --digest

# Normalizing 3 transcripts (Zoom format)…
# Processing: standup_mon.txt standup_wed.txt standup_fri.txt

Batch complete: 3/3 succeeded

Commitment check:
[MISSED] Bob — Deploy runbook (due May 14, 1 meeting elapsed)

Weekly Digest
Meetings: 3 · Action items: 11 · Decisions: 6
Open items by owner: Alice (3), Bob (2), Charlie (2)
Escalated risks: 2 (critical)

How it works

Four steps.
Zero ambiguity.

Messy transcript in. Structured accountability out — across every meeting this week.

Normalize

Strips Zoom, Teams, and Meet timestamps. Cleans speaker labels. Detects source format automatically

Classify

Claude identifies meeting type — standup, retrospective, exec review, incident review — with a confidence score

Extract

Decisions, action items with owners and deadlines, open questions, risk flags, and a type-aware follow-up email

Track

Cross-meeting commitment tracker flags missed items by severity. Weekly digest aggregates open items by owner across all sessions

Capabilities

Built for
accountability.

Every feature targets the gap between a meeting happening and work actually getting done.

Transcript Normalization

Strips Zoom [00:02:14], Teams -->, Meet (00:02) timestamps. Normalises email addresses and ALL_CAPS speaker labels. Reports source format and speaker count

Meeting Type Detection

Classifies each meeting as standup, retrospective, exec_review, sprint_planning, incident_review, 1-on-1, and more — with a confidence score. Type shapes the follow-up format

Batch Processing

Process an entire directory of transcripts in one command: python main.py batch --dir ./transcripts/. Failures are isolated — one bad file never stops the rest

Commitment Tracker

Compares action items across meetings. Flags missed commitments with severity (critical / moderate / low), counts meetings elapsed, and surfaces the highest-risk owner

Weekly Digest

Rule-based aggregation: open items by owner, escalated risks (deduplicated across sessions), carry-forward items. Single LLM call generates a ready-to-send digest email

Notion Integration

Push structured meeting output directly to a Notion database. Set NOTION_API_KEY and NOTION_DATABASE_ID, then pass --push-notion to any extraction command

Example

Chaos in.
Structure out.

Input — raw Zoom transcript

[00:01:12] ALICE_CHEN_EXT: Morning.
Redis still at 450ms — auth is
blocked.
[00:01:45] BOB_PATEL: I'll own the
fix by Wednesday.
[00:02:10] ALICE_CHEN_EXT: Deploy
needs Priya's sign-off. Flag it.
[00:02:30] BOB_PATEL: Should we
pull in the infra team?
[00:02:44] ALICE_CHEN_EXT: Not yet,
let's see how Bob's fix lands first.

Output — normalized + extracted

{
  "meeting_type": "standup",
  "meeting_type_confidence": 0.94,
  "action_items": [{
    "task":     "Fix Redis latency",
    "owner":    "Bob Patel",
    "deadline": "Wednesday EOD",
    "priority": "high"
  }],
  "risk_flags": [{
    "description": "Deploy blocked on
     Priya's approval",
    "severity": "critical"
  }],
  "formatted_follow_up": "🔴 Blockers\n
   - Deploy pending Priya sign-off\n
   🔄 In Progress\n- Redis fix (Bob)\n
   ✅ Done\n- Sprint kickoff"
}

Output schema fields

meeting_type

meeting_type_confidence

decisions[]

action_items[].task

action_items[].owner

action_items[].deadline

action_items[].priority

open_questions[]

risk_flags[].severity

formatted_follow_up

summary

Meeting to Action

Four steps.Zero ambiguity.

Built foraccountability.

Chaos in.Structure out.

Five more tools.Same standard.

Meeting to
Action

Four steps.
Zero ambiguity.

Built for
accountability.

Chaos in.
Structure out.

Five more tools.
Same standard.