went through the amazon data engineer loop for an L5 role on the consumer team in early 2026. the recruiting process was slow (three weeks from application to first call, six more weeks to close the loop) but the interviews themselves were pretty well-organized.
the loop: one phone screen with a hiring manager, then four virtual onsite rounds.
phone screen: sql on a shared doc. two questions. first was a basic aggregation with a GROUP BY and HAVING. second was a slowly-changing dimension type 2 implementation problem. you needed to write the INSERT/UPDATE logic to track history. if you don't know SCD type 2, look it up. it comes up.
onsite round 1: sql + data modeling. they gave me a raw event log and asked me to design a star schema for a reporting use case. then write a query against it. expect questions about normalization tradeoffs and why you'd denormalize for OLAP.
onsite round 2: systems / pipelines. design a pipeline to ingest clickstream data from a high-volume source. i talked through kinesis -> s3 -> glue -> redshift but they wanted specifics: how do you handle late-arriving data? what if a partition misses? how do you do idempotent writes? they weren't looking for THE right answer. they want to see you reason through failure modes.
onsite round 3: coding. this was more SWE-flavored. one medium LC problem (graph traversal) and one data-specific problem about parsing a nested JSON log structure in python. not as heavy as an SWE loop but you need to be fluent in python.
onsite round 4: leadership principles. customer obsession and ownership came up. have your LP stories ready and make sure they're specific and recent. they pushed on 'what was the impact' and 'how did you measure it' pretty hard.
i didn't get an offer this time. recruiter said it was a close call on the systems design round. going back to prep that harder. posting anyway because the details are useful.