Data Foundations and Audit Trails MOC
The substrate talks: without a governed data estate, every AI pilot rebuilds its own plumbing, and every agent multiplies the traceability obligations. The local-LLM investigation then asks what happens when the audit trail itself becomes the thing an AI reads.
Transcripts
- Shelly LaPointe - one governed dataset, many consumers; raw/curated/consumption zones; agent identity as a first-class access-control subject; the phased roadmap
- Vivekram Apparsundaram (Zifo) - RADAR dual-agent regulatory intelligence redlining impacted SOPs; a governance dashboard aggregating deviations, CAPAs, and training; checklist-driven Part 11 assessment
- Yari Fontebasso (Systera Solutions) - 113 local Ollama experiments on synthetic audit trails; the scenario ladder from Isolation Forest baselines to RAG; control versus capability; four-layer validation model
Evergreen notes
- GxP systems are optimized to be inspection-ready, not insight-ready
- Autonomous agents need their own auditable identities in regulated systems
- One governed dataset should serve many consumers
- Local models trade capability for control
- Absence problems are harder to detect than presence problems