Behavioral & Ownership: the practice test
Pushing back on a launch for reliability, the on-call story, the migration nobody wanted, working with researchers, the safety and mission rounds at the labs, and your view on where AI infrastructure is going. The rounds that decide between two technically equal candidates. This test drills exactly that: 10 easy, 10 medium and 10 hard questions, every one explained, every explanation linking into the worked material.
Sample questions, answered
The beats are built around what an infrastructure interviewer cannot get from a résumé: the instrument you read (per-rank timing, dmesg, a DCGM field, the p99 panel) and the two things you ruled out, the decision with its cost and the alternative you rejected, the metric that returned to baseline, and the rule, alert or automation that made the next occurrence a ticket instead of a page. Four minutes, a number in every beat. The heroic all-nighter version has none of those and fails.
A no is arithmetic plus a dated alternative; a yes has a condition; both are in the researcher's units. Saying 'the pod holds 1,024, three teams hold 60% under quota, your quota is 256, ten days at 1,024 is 246k GPU-hours against about 100k borrowable' is a conversation; 'the policy says' is a wall, and researchers route around walls. The three questions that find the need behind the ask are what the run produces and by when, what happens if it is late, and what they would cut with half the capacity.
Go deeper than the quiz
A practice test measures recall. The material it draws from teaches the reasoning: