How to Interview Data Engineers

Screen for judgement, not tool trivia

Data engineering is one of the scarcest markets we recruit in, and one of the easiest to hire badly. The failure mode is screening for a tool checklist — does this person know Spark, Airflow, Snowflake — instead of the judgement that keeps pipelines reliable and affordable. The tools change every couple of years; the underlying reasoning is what you are actually hiring.

The best data engineering interviews look like the work: realistic problems, scored on how someone thinks about data, scale and failure.

Test SQL and modelling properly

SQL fundamentals still matter and always will. Go beyond basic joins: window functions, aggregation, and the ability to reason about a query’s cost. Pair that with data modelling — dimensional modelling, and the trade-offs between a normalised warehouse and denormalised serving tables. Someone who models data well saves the whole organisation from slow, brittle analytics later.

Run data system design as a conversation

Give a concrete brief rather than a vague one: design a pipeline that ingests user events, validates and enriches them, stores the raw data, materialises serving tables and handles replay. The signal is in how they start. Strong candidates clarify before they solution — who produces the events, what latency is required, whether ordering matters, how long data is retained, and who consumes the output. Then they decompose the problem, estimate scale, choose patterns with explicit trade-offs and anticipate failure modes.

Watch the batch-versus-streaming instinct

One of the clearest senior signals is restraint. Default to batch unless there is a genuine latency requirement — roughly under five minutes. Batch is simpler to build, cheaper to run and easier to debug; a daily job that runs in twenty minutes and costs a few pounds usually beats a streaming pipeline that costs hundreds a day. A candidate who reaches for streaming on every problem is showing a lack of judgement, not sophistication.

Screen for the 2026 shift

The stack is evolving. Alongside SQL and Spark, roles increasingly involve orchestration (Airflow, Prefect) and LLM-based data pipelines, where the hard parts are rate limits, retries and cost budgets. If the role touches data or AI, screen for it honestly rather than bolting it on — our AI, ML and data recruitment page covers what to look for, and our Python developer hiring guide covers the language most of this is built in.

Move fast, and close

Data and AI talent is scarce and in demand, so a slow process loses people. Keep the loop short, give feedback quickly, and make sure every candidate deals with a real person.

In a market this scarce, the whole game is telling apart the people who can ship models and pipelines from the ones who can only demo them — and that is what our screen is built to find. If you are hiring, see how we hire data engineers and machine learning engineers, or talk to us and we will bring you people who can genuinely build.