went through mercury's data engineering loop last month. sharing notes because i couldn't find anything useful before my own interviews.
recruiter screen was pretty standard, 30 min. she asked about my background in pipelines, what tools i've used (dbt, airflow, spark etc.) and what kind of scale i'd worked at. no gotchas, just genuine curiosity.
then a take-home. they gave me a messy dataset and asked me to write SQL to answer a handful of business questions and then explain what i'd do if this needed to run daily at scale. i used window functions, CTEs, a join or two. the data was intentionally dirty, so part of the exercise was catching the nulls and duplicate rows. not hard, but it took me about 2 hours. they said 90 minutes. don't rush.
technical screen (video call, 60 min). live SQL questions first, similar to the take-home but fresh. the interviewer was an actual data engineer, not someone reading from a script. we also talked through a pipeline design problem: given event data coming from multiple sources at different cadences, design a system that makes it queryable within 15 minutes. i sketched a kafka -> flink -> warehouse approach, talked about late arrivals and deduplication. they pushed back a bit to see if i'd defend my choices or just fold. defend them.
final loop was 4 rounds back to back: pipeline design (deeper version of above), data modeling (star schema vs. wide tables, when you'd use each), a cross-functional collab round where they asked how i'd work with analytics engineers and product, and a values/culture fit.
things that seemed to matter: you need to be comfortable with dbt and modern warehouse tooling (they use snowflake i think) they care a lot about data quality: how you catch problems upstream, not just fix them downstream the culture round wasn't fluff. they specifically asked about times i pushed back on a stakeholder.
overall the loop felt well-designed. the interviewers had clearly talked to each other beforehand. offer was competitive for a series C fintech, though i ended up taking something else. happy to answer questions.