Jane Street · Primly Community

Jane Street data engineer interview: pipelines and SQL, what they actually care about

de_derek · 4 replies

went through the jane street data engineer interview loop earlier this year, nyc role, sharing what i saw since there's almost nothing out there on this specific track.

the recruiter screen was light, basically just timeline and interest. then they sent a take-home SQL exercise. not a toy problem. they gave me a schema representing something like a trading event log and asked me to do window functions, CTEs, some aggregation logic with edge cases around nulls and late-arriving rows. had a 72-hour window but it probably took 4-5 hours of real work if you want to do it well.

technical rounds (2 of them, both live): first was more SQL/data modeling. they had me design a schema for tracking order flow across markets, talk through partitioning strategy, indexes. real questions: "what happens to your query if this table has 10B rows and that join key isn't indexed?" second was pipeline design. draw out a streaming vs. batch tradeoff for some scenario. they referenced kafka and flink by name. i don't think you need to have used flink specifically but you need to understand the stateful stream processing concepts.

there was also a coding round, python. not leetcode hard, but you're writing real logic: parsing, transforming, handling some gnarly input format. they care a lot about clean code and will push back on shortcuts.

no behavioral rounds in the traditional sense. there was some conversation about projects and how i'd handled ambiguity, but it wasn't structured STAR-method stuff. more like normal technical conversation.

things that seemed to matter: understanding data freshness vs. latency tradeoffs, not just knowing the vocab being precise about what "at-least-once" vs "exactly-once" means in practice asking clarifying questions before designing anything. they rewarded "what's the acceptable staleness here" type thinking.

did not get the offer. debrief said coding was strong, system design was ok, SQL exercise had a subtle correctness issue i missed. honestly fair feedback.

if anyone else has done the DE loop here, would love to compare notes.

4 replies

infra_ines

that SQL exercise description is exactly what i'd expect from a place like JS. the late-arriving rows edge case is a classic. did they specify whether this was postgres or something proprietary? i've heard they use OCaml internally for a lot of things but wasn't sure if that bleeds into data tooling.

de_derek

sql was standard postgres-compatible syntax as far as i could tell. they didn't mention OCaml at all in the DE track but wouldn't surprise me if the underlying infra is different. they were definitely evaluating SQL logic not dialect trivia.

ds_dmitri

how long did the full loop take start to finish? and did they give comp numbers before or after the final round?

de_derek

about 5 weeks from first recruiter contact to debrief call. comp came up only at the very end after they knew they were making an offer (which in my case they weren't). no numbers shared proactively during the loop.