Turn risk into a decision
Define the use, consequence, evidence, accountable owner, and the condition that changes what may ship.
AI governance · applied AI systems
Gabriel Goodhart
I turn AI risk into a practical decision, an accountable owner, and evidence a cross-functional team can inspect.
I bring 25 years leading consequential work in regulated industries and direct experience building production AI systems since July 2023.
Across 35 systems, including 25 in production and three in the care sector, I build the intake, evaluation, release, monitoring, and change-control work that tells a team what can ship, what needs evidence, and what must stop.
No email wall on the resume. It opens in the browser and prints to PDF.
How decisions move
Five capabilities behind the work
Technical depth matters when it makes a consequential choice clearer, owned, and repeatable.
Make policy reach the system
Trace a material rule to its control, test, evidence, exception path, and next review trigger.
Keep expert authority explicit
Clinical, legal, privacy, security, compliance, technical, and operational judgments stay with their named owners.
Watch for action, not activity
A monitored signal names its bound, owner, first response, escalation path, and condition for reopening the decision.
Build controls teams can keep using
Risk-based review depth, reusable evidence, and automated completeness checks protect judgment from avoidable queue work.
Selected governance evidence
Three decisions that changed what could ship
Each case connects the risk, evidence, authority, release action, and condition that reopens the decision.
Verified experience
Stillwell
The policy had to reach the running system
Consent, identity, retention, and escalation rules became release conditions instead of advisory text.
Run the consent and escalation demoVerified experience
Plate Check
The medical feature stayed off
A product boundary and a missed evaluation threshold both changed what the system was allowed to do.
Read the Plate Check decisionVerified experience
agent-gate
Unsafe or unproven agent actions stop the run
An agent cannot turn an unsupported claim or unrecognized high-risk action into a successful result.
Read the agent-gate caseDecisions that held the line
Two times the evidence changed the release
Governance earns its place when a finding changes the product, the release, or the monitoring plan.
The policy had to reach the running system
A written safety rule was not enough. I traced each requirement to a control and test, named the gaps, and made policy violations release-blocking instead of advisory.
See the Stillwell control pathThe medical feature stayed off
A vision feature missed its labelled-data threshold. I kept it out of the product, retained the result, and treated the failed evaluation as a release decision rather than a prompt-tuning exercise.
See the Plate Check decisionNext step
See the governance work, open the résumé, or start a conversation.
The public portfolio stays concise. The full system catalogue remains available when you want the technical depth.