AI PMO Copilot · document-grounded · human-reviewed

AI PMO Copilot — PM AI Fluency Lab

Paste real project artifacts on the left. Generate status reports, risk summaries, stakeholder updates, and an action tracker on the right — every line traceable back to a source document, nothing invented.

Status Report

not generated

Risk Summary

not generated

Stakeholder Update

not generated

Action Tracker

not generated

What Changed

not generated

Document intake

source of truth

Project charter (free text)

Meeting notes — previous week

Meeting notes — current week (use "- [ ] task — Owner — due date" for actions)

Risk register CSV — id,description,severity(High/Medium/Low),owner,status

Stakeholder list CSV — name,role,interest(High/Medium/Low),channel

Console output

Run a generator to see grounded output here. Every fact is pulled from the documents in the intake panel — anything missing is labeled, not guessed.

Human review checklist

required before send

Every claim in the output traces to a specific source document — no invented facts. Items marked "Not found in source materials" have been checked with the team, not silently dropped. Risk severities and owners match the actual risk register, not an AI guess. Stakeholder tone and detail level fit that audience (exec vs. team). No confidential or personally identifiable data was pasted that shouldn't leave this system. A second person has skimmed the final output before it goes out.

Governance notes

responsible use

What this tool is good at

Turning scattered notes into a consistent first draft: pulling structure out of meeting notes, formatting a risk register into prose, and re-tuning the same facts for different audiences.

Where it can get things wrong

It has no judgment about which risks actually matter most this week, can't detect sarcasm or nuance in notes, and will faithfully reformat a mistake if the source document has one.

What should never be pasted here

Customer PII, credentials, unreleased financials, or anything under an NDA that this tool (or a real LLM backend) isn't cleared to process.

How to judge output quality

Check traceability first — can you point to the source line for each claim? Then check omissions — did anything important in the source get dropped silently?