PSETs are a low-stakes way to build muscle memory for these problems. Try as many attempts as you want before the deadline; your best attempt is what counts. Find a problem you got wrong? Retake it, resubmit, and you get full credit. The goal is to keep iterating until each concept clicks.
Code is how this material moves from notes to muscle memory. The colabs reinforce the lectures and the readings by letting you actually run the systems you're learning about: a real query plan, a real WAL trace, a real distributed job. That's where the intuition lives.
Build the portfolio you'll show at data interviews. Project 1 puts you in the cloud, querying gigabytes of live BigQuery data. Project 2 lets you ship either a data-science pipeline (Science track) or hand-build a database engine from scratch (Systems track); pick the one that maps to the role you want. Both projects push your AI workflow through its paces, so you'll feel firsthand where AI accelerates you and where you still have to know the systems yourself to ship something correct.