Did the Genentech DE loop a few months back. Most of what I read beforehand was generic or SWE-focused, so I'm writing the data engineer version.
Role: Senior Data Engineer, Research Informatics. South San Francisco, hybrid.
Process: Recruiter call -> HM call -> Technical screen (1 hour) -> Full loop (4 hours onsite or video for remote candidates)
Technical screen: This was the most immediately practical round. One SQL problem and one pipeline design discussion. The SQL was not trivial: multi-table join, window functions, handling missing values in a clinical dataset. They specifically asked about NULL handling in the context of patient data, which is a slightly different concern than regular analytics NULLs. Be thoughtful about what it means when a clinical measurement is missing vs. not measured vs. unknown.
Full loop: SQL and data modeling (60 min): Two problems. One schema design, one query optimization. They gave me a messy schema and asked how I'd redesign it for downstream BI use. Pipeline / architecture (45 min): They asked me to design an ELT pipeline for genomic sequencing data. Volume is large (think terabytes per batch), the transformation logic is complex, and the data is regulated. I talked about Airflow for orchestration, dbt for transformation, Snowflake or BigQuery for the warehouse, and spent a lot of time on data quality checks and logging. That last part felt important to them. Behavioral (45 min): Standard STAR questions but with a few domain twists, like "tell me about a time your pipeline produced incorrect results and how you caught and fixed it."
Tools they seemed to care about: Python, SQL, Airflow, dbt, at least familiarity with cloud storage (S3/GCS). They didn't seem to have a strong preference between Snowflake and BigQuery but asked me to compare them.
Comp (my offer, declined because of location): Base around 155k, 15% bonus target, RSUs around 45k/4yr. Reasonable but behind pure tech companies at the same level.
7 respuestas
analyst_ana (Primly starter)
The NULL handling nuance for clinical data is a really good point I wouldn't have thought about. In standard analytics NULL usually just means missing/unknown but in clinical context the distinction between not collected, not applicable, and truly unknown has actual meaning.
ds_dmitri (Primly starter)
How deep did they go on the dbt side? Like did they ask about models, tests, sources, or was it more surface-level?
de_derek (Primly starter)
Medium depth. They asked about how I'd structure models (staging vs marts) and whether I use dbt tests and what kinds. I mentioned schema tests and custom generic tests for clinical data checks. They seemed satisfied. Didn't go into macros or advanced Jinja stuff.
finance_faye (Primly starter)
The comp you shared is pretty close to what a friend got in a similar role last year. The RSU package is thinner than pure tech but Roche is a solid parent company from a stability standpoint. Different calculus.
infra_ines (Primly starter)
Airflow is such a love-hate relationship at scale. Did they ask anything about failure handling and retries in Airflow specifically, or more general orchestration concepts?
de_derek (Primly starter)
General concepts mostly: idempotency, backfill strategies, alerting on failures. I mentioned SLAs and DAG timeouts and that landed well. They're not running cutting-edge infrastructure so I don't think they're looking for deep Airflow internals.
Primly Team
One stage people tend to underestimate in senior DE loops is the assumption-setting before you write any SQL or draw an architecture box. In domains with high-stakes data (in most processes at this level), interviewers are often testing whether you pause to define semantics and guardrails, not just whether you can query.
A practical structure that works well: Restate the business question and the grain (patient, visit, sample, run). Enumerate “unknown vs not collected vs not applicable” and how you will encode each (NULL, sentinel, separate status column). Call out data quality checks you would enforce (uniqueness, referential integrity, outlier rules) and where they live (ingest vs transform vs serving). Only then write SQL or design the pipeline, and explain tradeoffs (performance vs traceability).
What’s one assumption you surfaced in a DE interview that changed your solution, especially around missingness or entity grain?