Private, verifiable AI
Private AI over your own documents, with every citation checked before an answer is shown.
I build and test AI for organizations that handle sensitive information. Answers cite their sources and are checked against them. Data stays on your own machines. Tools are stress-tested before anyone relies on them.
J.D., Washington attorney · IAPP AIGP · Tikor Consulting since 2003
What I do
Private AI over your own documents, with every citation checked before an answer is shown.
AI inventories, risk assessments and stress tests, so you know what your tools do before a client or regulator asks.
23 years of Oracle E-Business Suite and enterprise data work underneath it.
Evidence
98%+
citation accuracy, zero hallucinations: WA Legal AI on a 150-question attorney-reviewed test set
26 of 30 → 0
a frontier AI agent crossed an authority boundary in 26 of 30 held-out runs; one sentence of task framing took it to zero
$4M to $40B
the range of client organizations since 2003
Work
Legal AI
In-house prototype
An air-gapped, on-premises research assistant for Washington State law. It answers questions over about 271,000 authorities (statutes, regulations, court rules and case law) and checks every citation against the retrieved sources before an answer is shown.
AI evaluation
Research
A harness that tests how AI agents behave under pressure.
Cloud migration
Fortune 100
Led the migration of relational databases from a private data center to Amazon Web Services.
Writing
Applied AI
The hype says "AI everywhere." The returns say something narrower. Here's where LLMs earn their keep inside an ERP — and where they quietly burn budget.
ERP & M&A
Before you migrate — or buy a company that runs on legacy ERP — a short, honest readiness assessment saves far more than it costs. What to actually look at.
More essays on AI governance, technology and policy at The Correction Path →
A fixed-scope review of the AI tools your team already uses, tested on your own material. You get written findings and a control plan.