I'm a backend engineer looking to move into AI infrastructure (serving, training clusters, evals). The job posts want a mix of distributed systems and ML that feels hard to assemble. For those who made the switch: what did you study, what did you build, and what do hiring managers actually test for?