Starbucks · Primly Community

Starbucks data engineer interview, pipelines and SQL: what actually came up

ds_dmitri · 4 replies

Just finished the Starbucks Technology data engineering loop in Seattle. Four rounds, all remote, took about five weeks from recruiter reach-out to verbal offer.

Here's what they actually tested:

SQL. A lot of it. Two dedicated SQL rounds, not just a warm-up question. One focused on window functions, lag/lead, calculating things like running totals and customer purchase frequency across multiple stores. The other was a messier data-quality scenario: given a table that had duplicate rows and some nulls in timestamp columns, write a query to deduplicate and explain your logic. They wanted to hear me think out loud, not just produce syntax.

Pipeline design. One 45-minute session that was basically a system design round for data. The prompt was something like: design a pipeline that ingests real-time loyalty card transactions across 16,000+ stores and makes them queryable by the analytics team within 15 minutes. I talked through Kafka for streaming ingest, Spark Structured Streaming for transformation, landing to a data lake in Parquet, then a Databricks SQL endpoint for analysts. The interviewer asked about late-arriving events and how I'd handle schema evolution. Both fair questions.

Behavioral. More than I expected for a data eng role. Two dedicated behavioral sessions. They really pressed on cross-functional collaboration, specifically working with business stakeholders who don't know what a DAG is. Makes sense given the company. The data team serves product, marketing, store ops, and supply chain simultaneously.

Tools they mentioned. Databricks, Azure Data Factory, dbt, Azure Synapse. If you're coming from an AWS-first background, worth brushing up on the Azure equivalents. Their data platform is pretty Azure-heavy.

Level. I interviewed at the senior data engineer level (roughly equivalent to L5 at bigger tech companies). The SQL bar felt higher than I expected from a retail brand, honestly. They've clearly invested in the data function.

Happy to answer questions about the SQL problems or pipeline design round specifically.

4 replies

analyst_ana

Thank you for the detail on the SQL rounds. Did they give you sample data to write against, or was it more whiteboard-style where you describe the query?

ae_andre

They used a shared screen with a SQL editor connected to a sample schema. Not a real database, but you could actually run queries and see results. Made it feel more real than whiteboard SQL, which I appreciated.

sdr_sky

The loyalty card transaction pipeline prompt is exactly the kind of thing I'd expect from a company with that volume of point-of-sale data. Good that you covered late-arriving events, that's usually where interviewers separate people who've operated pipelines vs. just designed them on paper.

content_cole

Azure-heavy stack checks out. Most legacy retail companies that weren't cloud-native from the start ended up on Azure because of Microsoft enterprise agreements. Worth knowing before you go in.