The process was a recruiter and tech screen on CoderPad with advanced SQL and some Python data manipulation. The onsite was a five-round loop: data modeling, pipeline design, Python coding, advanced SQL, and a Googleyness and behavioral chat.

The pipeline design round is the absolute filter in my opinion. They throw a massive question at you, like designing a real-time analytics pipeline for an ad-click tracking system, and you have to map out the batch versus streaming architecture, idempotency, and how to handle late-arriving data from scratch.

The modeling round was all about star schemas, handling slowly changing dimensions, and optimizing tables. It felt very practical. It is more about how you structure data at a massive scale and handle edge cases.

For preparation, I focused on the Kimball methodology and practiced streaming architectures. I did a few mocks on Prepfully with Google Data Engineers specifically for the pipeline design round, and that helped me a lot. I skimmed through DataLemur for questions and practice, and of course Blind and Reddit for recent experiences.

Design a data pipeline to process clickstream events from millions of users in real time.