Vercel · Primly Community

Vercel data engineer interview: pipelines and SQL, what they actually tested

de_derek · 5 replies

Went through the Vercel DE loop earlier this year. Four rounds total. Wanted to write this up because most Vercel content focuses on frontend SWE and there's basically nothing on the data side.

Recruiter screen was pretty standard: background, why Vercel, general curiosity about their data infrastructure. About 30 minutes.

Technical screen: one SQL question and one Python/data modeling question. The SQL was a multi-table join with a window function, nothing sadistic, but you had to explain your reasoning out loud. Think: given a table of deployment events with timestamps, find the average time between builds per project for the last 30 days, exclude projects with fewer than 5 builds. Classic DE interview SQL.

The pipeline design round was the most interesting. They gave me a vague scenario around ingesting high-volume build telemetry data into something queryable by the analytics team. No specific stack prescribed. I talked through Kafka for streaming ingestion, landed on a Lakehouse-ish pattern (Bronze/Silver/Gold), mentioned dbt for transformation. They pushed hard on data quality: how do you catch schema drift, how do you handle late-arriving events. The conversation stayed technical and felt like a real engineering discussion.

Fourth round was behavioral. Nothing unusual: a time you had to handle an incident with bad data, how you've worked with downstream consumers, disagreement with a stakeholder.

What I didn't get asked: no system design in the traditional SWE sense, no distributed systems deep-dive. The focus was clearly on the data layer and on communication.

One thing to know: Vercel's data team is small. The DE role felt more like a founding-team-adjacent hire than a cog in a big analytics org. They seemed to care whether you can think independently and own an area end to end.

Timeline for me: recruiter screen to offer was about 5 weeks. Final debrief was quick, same-week turnaround.

Happy to answer specifics below.

5 replies

content_cole

Really helpful, thank you. Did they ask anything around data warehouse choice? Like did they probe you on Snowflake vs BigQuery vs Redshift tradeoffs, or was it more architecture-pattern agnostic?

pivot_pat

Mostly pattern-agnostic. I mentioned BigQuery briefly but they didn't push on vendor specifics. The design question was more about the flow and the guarantees you'd want, not "name the service." That said, knowing your way around at least one columnar warehouse is table stakes.

analyst_ana

Curious whether the SQL round was live coding in a shared editor or whiteboard style? I always freeze more on one vs the other.

mobile_mara

Live coding in a shared editor with execution. Not whiteboard. You could run your query and see results against a small sample dataset. That actually helped a lot.

quietquit_quincy

Five weeks start to finish sounds pretty fast honestly. Were you also told a 5-week timeline upfront or did it just happen to land there?