Agentic workflows that generate audit-ready validation evidence for AI systems in GxP environments.
Regulators are moving fast — EU Annex 22 (final expected 2026) restricts generative AI to non-critical GMP applications under human oversight, and the ISPE GAMP 5 AI Guide defines the validation lifecycle. This framework turns those requirements into executable workflows:
- Applicability Assessment — an agent walks your AI use case through the Annex 22 decision tree: critical vs. non-critical, static vs. generative, and the validation rigor each tier demands.
- Human-Oversight Evidence Engine — for non-critical LLM applications (the most common enterprise case), generates the review trails, output-validation records, and sign-off chains that regulators mandate.
- Test-to-Clause Mapping — ingests outputs from LLM security and quality tools (spikee, DeepEval, RAGAS) and maps results to GAMP 5 / Annex 22 requirements, producing audit-ready evidence packages.
Scope & responsibility: This framework supports validation activities for the AI layer of your system (Annex 22 territory) and integrates with your existing CSV process (Annex 11 territory) — it does not replace it. Compliance determinations always remain with your organization's Quality unit; GMP responsibility remains with the manufacturer. This tool makes those determinations evidence-based and fast.