Hi all,
Senior CSV/CSA Engineer here (9+ years in GMP-regulated pharma/biologics validation — GAMP5, 21 CFR Part 11, EU Annex 11) trying to go deeper on AI/ML validation in GxP environments, and would love to connect with anyone actually working in this space.
What I’ve been digging into so far:
- GAMP5 2nd Edition Appendix D11 (AI/ML lifecycle: concept → project → operation)
- ISPE’s GAMP AI Guide (published July 2025)
- FDA’s finalized CSA guidance (Sept 2025) and how it explicitly extends to AI tools in production/quality systems
- EU GMP Annex 22 (draft, AI in GMP manufacturing) — particularly the static/locked model vs. dynamic model distinction for GMP-critical use
I’ve built a small proof-of-concept myself — a risk-based review tool that keeps AI strictly to a “draft narrative” role while all pass/fail determinations stay deterministic and human-signed — but I’d really value input from anyone who’s:
- Actually validated a production AI/ML system in a GxP environment
- Used AI-assisted features in platforms like ValGenesis VLMS or Kneat
- Has thoughts on how model drift/retraining should be handled under change control in practice, not just on paper
Happy to share what I’ve built in exchange for a reality check from people further along on this than I am. Open to DMs too if easier.
Thanks in advance!