AI governance · Agent infrastructure · AI security
Put AI to work without losing control of your data.
I help engineering and security teams decide how AI agents are built, what they may touch, and how to prove it. Advice from a decade of large-scale infrastructure at Google and agentic AI architecture at NVIDIA.
What governed AI looks like
Every agent request, checked before it touches your data.
Agent asks
An AI agent requests data or an action.
Identity
It acts as the real user, not a shared account.
Policy
Rules decide: allow, mask or block.
Data
Only permitted data returns. Sensitive values are masked.
Audit
Every decision is recorded and can be reviewed.
The evidence trail
Allowed, masked or blocked, and always recorded.
Governance only counts if you can show what happened. A good design logs each request, who it ran as, which rule applied, and what was withheld, so a security review is a read, not an investigation.
Where I help
Three places AI projects get stuck.
AI governance review
What your agents can see, do and send out. A written review with ranked fixes you can act on.
Agent infrastructure design
Architecture for agents that run on your systems, with your models, your policy and your audit trail.
Secure retrieval and access control
Search that answers only from what each person is allowed to see, so AI cannot leak across teams.
Who this is for
Teams that hold data they cannot lose.
- Banks, insurers, healthcare and other regulated teams.
- Security and platform leads asked to approve AI tools they did not choose.
- Engineering teams building their first internal agents.
I also run Sparcle, which builds AI governance software. I will tell you whenever a recommendation touches it. Most of my advice does not depend on any product.
How it works
Simple, written down, and fixed in scope.
Talk
A free 15 minute call. You describe the problem, I tell you honestly if I can help.
Scope
A working session or a fixed-scope engagement, agreed in writing before any work starts.
Deliver
Clear written findings and a plan, not a slide deck of generalities.
Training
Teach your team to build and secure agents.
Workshops and short courses with hands-on labs, built for engineering and security teams.
- AI agent security for engineers
- Building governed AI agents
- AI governance briefing for security and risk leaders
Writing
Notes, research and guides.
- Guide
Five questions to ask before an AI agent touches your dataA one-hour checklist for security and platform leads who are asked to approve an AI agent they did not build.
Start here
Start with a conversation.
Fifteen minutes, free, no preparation needed. Tell me what you are trying to do.