just finished a Zoom DE interview loop last week for a senior data engineer role. this is a bit niche so figured I'd document it since there's basically nothing online about what their DE process looks like.
process overview: recruiter screen, HM screen, then 4-round virtual onsite. mine was fully remote (I'm not in the Bay Area).
round 1: SQL. 45 min. table structure was Zoom meetings, users, and a log of meeting events (join, leave, audio mute, etc.). questions: find the average meeting duration by host department for meetings with more than 5 participants identify meetings where at least one participant joined more than 10 minutes late (requires time diff calculation) the hard one: build a 'user activity funnel' from raw event logs. messy data, out-of-order events, null handling required.
I was comfortable with all of these but the funnel query took me close to the limit. SQL at this level rewards knowing when to use CTEs vs subqueries for readability.
round 2: pipeline design. system design but DE flavor. they gave a scenario: Zoom wants to ingest clickstream data from all clients (desktop, mobile, web) in near-real-time for experimentation and analytics. design the end-to-end pipeline.
I talked through Kafka for event streaming, Spark Structured Streaming for processing, landing in a data lakehouse (Delta Lake / Iceberg). they cared a lot about schema evolution, late-arriving data, and exactly-once semantics. deduplification strategy was a big sub-topic.
round 3: coding. more like a data transformation coding problem in Python. manipulate a nested JSON event log, compute derived metrics. not algorithmic, practical.
round 4: behavioral. standard. what's a pipeline you built that broke in production and how did you fix it. have a good war story ready.
impressions: the process felt well-run and the interviewers knew their stuff. no gotcha questions. they want people who can build reliable, scalable data infrastructure, not just write clever SQL.