PayPal · Primly Community

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

de_derek · 4 replies

Just wrapped the PayPal DE loop last month, senior-level, San Jose. Took about 4 weeks from recruiter call to offer. Sharing because I couldn't find much recent data on this.

The loop was 5 rounds: one phone screen (SQL and system design lite), then four onsites.

SQL: heavier than I expected. They wanted window functions, CTEs, and one question specifically around deduplication logic in a payment event stream. Not LeetCode-style joins, more like: here's a messy transactions table, find customers with duplicate charges within a 30-minute window. Wrote it cold on the whiteboard. Practice actual analytical SQL, not just SELECT * FROM.

Pipeline design: two rounds were entirely pipeline/architecture. One was about ingesting high-volume payment events at scale (think Kafka, schema evolution, exactly-once semantics). The other was building a fraud signal pipeline with latency constraints. I'm not sure they expect you to know PayPal's internal stack, but you should be able to talk trade-offs between batch vs. streaming, idempotency in pipelines, and monitoring approaches.

Tools that came up: Spark, Flink (mostly conceptual), Airflow for orchestration, Hive. The actual question was 'what would you use and why' not 'demonstrate mastery of this specific tool.'

Data modeling: one round was data modeling for a hypothetical merchant analytics product. Dimensional modeling (fact/dim tables), slowly changing dimensions, the usual. Know your Kimball basics.

Behavioral: two rounds mixed behavioral in. Standard STAR stuff. They asked about a time you caught a data quality issue before it hit a dashboard, and what you did when a pipeline you owned went down during a high-traffic event. Fintech-flavored, but not unusual.

Offer was around $210k TC for L5 equivalent, San Jose. Stock portion was relatively light compared to big tech but base was solid.

Happy to answer follow-ups. The SQL round is no joke, do not walk in unprepared for the analytical side.

4 replies

analyst_ana

the deduplication question is such a classic fintech problem. did they expect a specific solution or was it more about talking through your approach? i always get nervous when they say 'write it cold on the whiteboard' for sql

de_derek

they wanted working SQL. not pseudo-code. i stumbled on a syntax detail and they let me finish without interrupting, so they're not looking for perfection, but the logic had to be right. i'd say 70% of credit was on the approach, 30% on getting the syntax clean.

content_cole

the 'exactly-once semantics' question in pipeline design always trips people up. it's one of those things where the interviewer usually knows there's no perfect answer, they just want to see if you understand the trade-offs (at-least-once + idempotent consumers, vs. transactions, etc.). did they go deep on that or keep it high-level?

sre_sol

I did a PayPal SRE loop about 8 months ago and the pipeline design round bled into reliability territory for me too. they really care about 'what happens when X fails' across the entire system, not just the happy path