CMC Knowledge System
Makes templates and checklists from the lessons of past submissions and audits.
Interactive walkthroughThis is a scripted walkthrough with synthetic data. It shows how the tool works without real program data.
Ask the knowledge base
Select a question to see what the team learned before.
Best match
Playbook for pre-approval inspection readiness
Used on 4 programs · 9 lessons added
- Link each commitment in the filing to the site record that proves it
- Do a practice walkthrough of the commercial line with the people who will present
- Close or justify the open deviations and CAPAs that apply to the filed process
- Prepare a short storyboard for each high-risk topic, such as data integrity
- Put a document runner in the back room, and answer each request in 30 minutes or less
Learned from
Audit finding, 2023Submission retrospective, BLA 2024Mock inspection debrief, 2025Why I built it
Audit findings, submission retrospectives, and reviewer comments often showed the same lessons again. Each time a new person joined, the team rebuilt the same governance logic from the start.
This tool started from my work at Novavax. See the role →
What it does
- Makes repeat work standard, but leaves room for judgment.
- Keeps the review logic and lessons that usually disappear between programs.
- Gives usable materials for onboarding and for repeat plans.
- Gives better continuity across filing and lifecycle work.
- What goes in
- Logic from past submissions, review patterns, lessons learned, reusable technical content, and working documents.
- What comes out
- Templates, playbooks, checklists, reusable one-pagers, onboarding materials, and shortcuts for governance tasks.
- What it changes
- Improves continuity, onboarding speed, and the quality of repeat work across programs and lifecycle activities.
How it stays inside the rules
- What it is for
- The tool finds insights from past submissions, audit findings, and lessons learned. It puts them into playbooks and templates that program teams and new team members can use again.
- What it never does
- This is a knowledge layer, outside the GMP boundary. Some content comes from audit findings or inspection outcomes. For that content, access control and correct attribution are more important than model performance.
- Who decides
- A precedent that the tool finds is only a starting point for a decision. It has a citation to its source artifact. It never replaces a new assessment against the current guidance.
- Risk level
- Low under FDA's draft AI framework (how much the output influences a decision × how serious that decision is)
- Guidance it follows
- FDA draft guidance · EMA reflection paper
What already exists
- SharePoint, Veeva Vault QualityDocs: Document storage, control, and retrieval
- Confluence and internal wikis: Written process and team knowledge
- The gap it targets
- Document systems keep the documents but lose the reasoning. A new PM asks, "Why did we answer that health authority question that way?" In my experience, no system held the answer, so each program had to learn it again.
- What it is not
- It is not a content repository. It is a retrieval layer for the reasoning behind the content.
Do you build tools like this?
I am glad to compare notes: what worked, what a reviewer did not accept, and where AI should stay out of the decision.