Evidence
What source, confidence, freshness, and caveat should accompany an AI output before a professional relies on it?
Pinavia research
Research for AI work that has to be explained, reviewed, and defended.
Pinavia studies the operational layer between institutional evidence and AI output: provenance, authority, review design, and proof. We do not treat a polished answer as evidence of a safe workflow.
Research agenda
Authority
Which verbs can the system prepare, and which must always remain a named human action?
Workflow
Where does AI reduce review friction without erasing the approvals, records, and accountability the workflow requires?
Regulated pilot patterns
The cases we are designed to test, without pretending they are finished case studies.
01
Regulatory readiness
A compliance team scopes one jurisdiction and licence objective, identifies missing evidence, and prepares a reviewer-owned pack. Meridian does not file or certify.
Pilot pattern, not client claim02
Board continuity
A board team compares a new pack with a stable prior baseline, highlights material movement, and routes director questions into human-owned decisions. Quorum does not approve or sign.
Pilot pattern, not client claim03
Diligence coverage
A deal team maps governed evidence to a checklist, surfaces gaps and red flags, and prepares material for a named advisor and committee. Vantage does not recommend investment.
Pilot pattern, not client claimHow we evaluate a pilot
- 01
Start with a real operating question, not a model benchmark.
- 02
Use a narrow evidence pack with a named source owner and a documented sensitivity boundary.
- 03
Measure evidence coverage, review effort, unresolved gaps, and the quality of the human handoff.
- 04
Publish pilot findings as observed workflow evidence, not as anonymous ROI claims or synthetic case studies.
Bring a real question
Run a diagnostic before making a broad AI claim.
We start with one workflow, one sponsor, an evidence pack, and a visible human boundary.