went through the full xAI data engineer loop in january 2026. five rounds total. sharing notes while it's fresh because i couldn't find anything useful when i was prepping.
recruiter screen (30 min) standard. they asked about scale i'd worked at, specific pipeline tools, and whether i was comfortable with ambiguity. the recruiter was pretty direct about the culture: fast-moving, no hand-holding, you own your stuff. that part wasn't BS.
technical phone screen (60 min) two parts: SQL and system design lite. the SQL was harder than i expected. not just joins and aggregations. they gave me a messy schema with multiple fact tables and asked me to write a query that tracked user sessions across multiple event types and flagged anomalies in the sequence. window functions, CTEs, the works. i'd put it at leetcode-medium-hard in terms of SQL difficulty.
the system design part was about designing a real-time ingestion pipeline for a high-volume event stream. think kafka-style architecture. they asked about partitioning strategy, consumer group design, exactly-once semantics. i know this stuff cold so it was fine, but people who've only done batch pipelines might get caught flat-footed here.
onsite (4 rounds, all virtual) coding round 1: more SQL, plus some python for data transformation. asked me to parse semi-structured JSON logs and output normalized tabular data. nothing wild. coding round 2: full system design. design a feature store for ML models. latency requirements, freshness tradeoffs, read vs write path. one of the better system design prompts i've seen. data modeling round: given a verbal description of a business domain, design a schema. they want normalization decisions justified, not just the schema. behavioral round: standard STAR stuff. tell me about a time a pipeline you owned failed in prod. what you built, what broke, what you changed. have a real story for this.
things that mattered: they care about end-to-end ownership. they want to see that you understand the whole stack from ingestion to consumption, not just one piece. knowing airflow or dbt isn't enough; they want to know you've debugged the thing at 3am and understand why it failed.
level was roughly senior DE. took about 3.5 weeks from first screen to offer. offer was competitive but i ended up not taking it.