Question 1
Design the pipeline that feeds the decision this posting cares about. What is the grain and the freshness?
Approach. Source, grain, and what a late or partial run should do. Say who notices.
Data engineer interview questions from a job description
See the modeling, pipeline, and quality questions a data engineer job description is likely to produce.
Matched in your browser from the requirements in the posting and Luna’s interview-question hub. These are practice prompts, not a claim that this wording will be asked.
Requirements in the posting
Data Engineer
Build the warehouse and pipelines other teams trust.
You will:
- Model data in SQL in the warehouse, and document the grain
- Orchestrate pipelines with Airflow and dbt, including backfills that do not quietly rewrite history
- Use Python and Spark where the volume requires it
- Set freshness expectations and say what a late or partial run should do
- When bad data reaches a reader, name the check and the owner who gets it
We will ask what a reader should do when today’s run did not land.
Build the warehouse and pipelines other teams trust.
You will:
- Model data in SQL in the warehouse, and document the grain
- Orchestrate pipelines with Airflow and dbt, including backfills that do not quietly rewrite history
- Use Python and Spark where the volume requires it
- Set freshness expectations and say what a late or partial run should do
- When bad data reaches a reader, name the check and the owner who gets it
We will ask what a reader should do when today’s run did not land.
Likely questions
System design
Data pipelines
Data Engineer
Question 2
How do you backfill without quietly rewriting history people already used?
Approach. Say what is immutable, what you version, and how a reader knows which run they are on.
Question 3
Tell me about bad data that reached a dashboard or a model. How did you stop the next one?
Approach. Name the check you added and the owner who gets it when the check fails.
Coding
Python
Use Python and Spark where the volume requires it
Question 4
How would you structure a Python module for the core task in this posting so someone else can change it?
Approach. Name the function boundary and the test you would write before adding libraries.
Question 5
Tell me about a time a Python script or service was correct and still too slow.
Approach. Say what you measured, the change, and the cost you left in place.
Question 6
What belongs in a notebook, and what has to leave the notebook before this team can trust it?
Approach. Draw the line at anything a customer or a pipeline reruns. Say how you would test that piece.
Coding
SQL and data modeling
Model data in SQL in the warehouse, and document the grain
Question 7
What tables would you need for the core object in this posting, and what is the grain of each?
Approach. Name the entity, the key, and one query you must keep fast.
Question 8
Write the shape of a query this team would run every day. Where would it get slow?
Approach. Say the join, the filter, and the index or rewrite you would try first.
Question 9
Tell me about a migration that changed a table people already depended on.
Approach. Two phases and a rollback. Say who felt it if you were wrong.
Behavioral
How you work
Build the warehouse and pipelines other teams trust.
Question 10
Which line in this posting is the closest match to work you have already done?
Approach. Read the line back, then spend the answer on what you did and the result, not on your title.
What to prepare
- Be ready to describe one pipeline: source, grain, freshness, and what happens when it is late.
- Solve one small task in Python while saying the cost of the obvious version.
- Practice one join out loud: the grain of the result, then the filters.
- Write the requirement you cannot yet support with a story, and one way you would learn it.
- Mark the three requirements you cannot yet support with a story.
- Say each answer out loud once, in under two minutes.
Data engineer postings interview grain, freshness, and backfills. Warehouse and SQL bullets become the model. Airflow, Spark, or dbt bullets become the run: what a late or partial job should do, and who notices. Python is often the transformation screen. “Reliable datasets” is the bad-data story.
Stakeholders appear as “a dashboard was wrong.” Your answer needs the check you added, not only the repair query.
Be ready to describe one pipeline from source to reader, including the immutable part. The sample is a warehouse role.
Sample job description
The posting below is already in the tool. The questions are the ones its requirements point to. Replace it with a posting you are actually interviewing for.
Related: Netflix system design questions · SQL interview pacing · Stripe interview questions