Use these questions to interview Data Engineer candidates at {{Company name}}. The role sits between software engineering and analytics, so probe both directions: engineering discipline (testing, idempotency, operations) and data judgment (modeling, quality, serving consumers well). Ask for real pipelines, real volumes, and real failure stories throughout.

Pipeline design and orchestration

• Walk me through the most complex pipeline you own end to end: sources, transformations, destinations, and schedule.
• How do you make a pipeline idempotent, so re-running yesterday is always safe? Show me a real design.
• Tell me about handling late-arriving data. What did your solution look like?
• How do you decide between batch and streaming for a new use case? Give a call you made and why.
• Design live: ingest a third-party API with rate limits, occasional downtime, and schema changes. Sketch the pipeline.
• What does backfilling two years of data look like in your current stack — how do you run it without disrupting daily loads?

What good looks like: strong candidates design for failure as the default — idempotent writes, checkpoints and retries, explicit late-data strategy — and can quantify their pipelines with volumes and SLAs. Weak candidates describe happy-path DAGs and treat re-runs, backfills, and schema drift as annoying exceptions rather than the job itself.

Data modeling and warehousing

• Walk me through a warehouse model you built — layers, naming, and how consumers find things.
• How do you model slowly changing dimensions? Give a case where history actually mattered.
• Tell me about a modeling decision that made analysts' lives visibly easier — or one that backfired.
• How do you handle the same entity arriving from three source systems with conflicting attributes?
• Where do you draw the line between transformation in the warehouse and logic in BI tools?
• What is your approach to testing transformations — schema tests, data tests, unit tests?

Performance and cost

• Tell me about a pipeline or query whose runtime you cut dramatically. What was the bottleneck?
• How have you reduced warehouse or compute spend? Give real numbers.
• How do you choose partitioning and clustering keys? Walk me through a real decision.
• When is it right to pre-aggregate versus compute on demand?
• What monitoring tells you a pipeline is getting slower or more expensive before users notice?

Data quality and reliability

• Tell me about the worst data incident you have dealt with — bad data reaching consumers. What was the root cause, and what changed?
• What data-quality checks do you build in by default, and where do they run?
• How do you handle an upstream schema change that would silently break your pipeline?
• How do you communicate a data outage to analysts and executives who rely on the dashboards?
• What does data lineage look like in your stack, and when did it save you?
• How do you decide which datasets deserve SLAs, and what do those SLAs say?

What good looks like: the best candidates treat quality as an engineering system — layered checks with owners and alerts, contracts or schema enforcement at ingestion, and an incident story with a concrete prevention that followed. Distrust candidates who have never shipped bad data; the honest ones have, and they know exactly what they changed afterwards.

Collaboration and consumers

• Who consumed your data — analysts, ML, product — and how did their needs shape your design?
• Tell me about a request you pushed back on because the pipeline would be unmaintainable.
• How do you onboard an analyst onto datasets you own?
• Describe working with software engineers to get better event data emitted at the source.
• How do you document datasets so people trust and correctly use them?