Anduril · Primly Community

Anduril data engineer interview: pipelines and SQL were both heavy, here's the breakdown

de_derek · 5 replies

Went through the Anduril data engineer interview loop about six weeks ago for a senior DE role out of their Costa Mesa office. Made it to offer (accepted, started last month). Sharing notes because the pipeline side is genuinely different from what I'd seen at more typical product companies.

Recruiter screen was pretty standard: 30 minutes, scope of role, visa question (I'm not on a visa but they asked upfront, which I actually appreciated). Then a take-home SQL challenge. They sent a schema for what looked like a simplified logistics/supply-chain dataset, three tables, and gave 72 hours to answer five questions ranging from window functions to some gnarly recursive CTE work. Not LeetCode-style, actual messy data with nulls and edge cases you had to handle. Budget about four hours serious work.

Technical phone screen: One hour with a senior DE on the team. First 20 minutes was drilling into my take-home answers, including why I chose one join order over another. Then 40 minutes of pipeline design: they gave me a scenario (basically: you have drone telemetry streaming in at high volume, design the ingestion layer). I talked through Kafka, partitioning strategy, schema registry with Avro vs Protobuf, and they pushed back specifically on failure modes. How does the consumer recover if it crashes mid-batch. What does exactly-once delivery actually mean here. Good signals: they care about reliability engineering, not just "I know Spark."

Onsite (4 rounds): Deep pipeline design round: more of the above but longer. Had to whiteboard an end-to-end data architecture including storage tiers, partitioning, and a backfill strategy. SQL and data modeling: they gave me a fuzzy requirements doc and I had to propose a schema and then write queries against it. No IDE, but they were reasonable about syntax. Cross-functional collab round: worked through a scenario where a PM wanted a metric defined in a way that was technically ambiguous. How do I clarify it, what data's missing, how do I push back. Mission and culture: not a soft skills formality. They actually probed how I think about the end use case of the systems I build.

Comp for senior: base came in at 175k, equity was RSUs with a 4-year vest. Total package landed around 230k depending on equity refresh. Southern California CoL math is fine.

Main prep advice: get really sharp on Kafka internals, window functions, and data modeling from ambiguous requirements. Leetcode is mostly irrelevant here.

5 replies

ds_dmitri

The take-home SQL design matches what a friend described earlier this year. He said the recursive CTE question specifically was more about whether you understood when to use it vs when a self-join is cheaper. Did you get docked points anywhere on the take-home that came up in debrief?

de_derek

Yeah actually. I over-indexed on readability and used a CTE chain that was technically slower on a large table. The interviewer flagged it but framed it as "in production we'd want to verify this" not as a hard knock. I think explaining the tradeoff and saying you'd benchmark it saved me. Don't just write the "clean" version, narrate your assumptions.

backend_bekah

The exactly-once semantics push sounds very deliberate. Most interviewers ask about it theoretically; if they're grilling on actual consumer recovery semantics they probably have war stories from prod. How long was the full loop start to finish?

visa_vik

Did asking about visa upfront affect anything or was it purely logistical? I'm on H1B and that question always gives me a mini heart attack.

de_derek

Pure logistics from what I could tell. They sponsor, they just want to know timelines early. Didn't feel like a filter. But Costa Mesa is not SF/NYC, so factor in that visa transfer timelines and location might interact for some people.