Owns the pipelines, deployment, monitoring and cost of models in production.
Why this one is hard to judge
The work is invisible when it is done well, so the evidence is in incidents and spend, not in features.
What to ask for
Ask what breaks first when model traffic doubles overnight, and how they would know.
Ask for a rollback they have actually performed, and how long it took.
Ask how they track which model version produced which prediction, months later.
Ask what they would cut first if the platform budget was halved.
The mistake most hirers make
Hirers focus on the tools listed on the CV rather than what happens when something fails under pressure. A platform built by someone who has never had to roll back a bad model often looks fine until it is not. Ask about the failure, not the stack.
What good looks like after 90 days
A rollback or recovery process that has actually been tested, not just documented. Clear visibility into which model version is serving which prediction. One piece of manual, error-prone work removed from the team's routine.
How we assess it
Against a rubric that is published in full, on evidence the practitioner supplies and a reviewer checks. Where something has not been verified, the profile says so.