Role
Data engineer
Data engineering listings read as a tool list and screen as a judgment test. The tools change every two years; the thing a team is buying is somebody who has been woken up by a pipeline that failed silently and designed so it could not happen again.
A data engineer builds the path from where data is produced to where somebody decides something with it, and is responsible for it still being true at the far end. The work is mostly about failure: late data, duplicated rows, a schema somebody changed upstream without telling anyone.
What most listings require
SQL
What it screens for: Not syntax. Whether you can read a query plan and know why a join is slow.
A sentence that proves it"I rewrote our daily revenue query from three nested subqueries into a window function; it went from eleven minutes to twenty seconds."
A pipeline or orchestration tool
What it screens for: Whether you have run scheduled work that other people depended on being correct.
A sentence that proves it"I have owned our Airflow instance since 2023 — about forty DAGs, including the one finance closes the month on."
Data modeling
What it screens for: Whether you have made a modeling decision and lived with it.
A sentence that proves it"I designed our events schema; the decision to keep raw events immutable and build aggregates on top is what let us recompute three months of metrics after we found the timezone bug."
Python
What it screens for: The glue. Whether you can write maintainable code, not just scripts.
A sentence that proves it"I wrote the ingestion library our four pipelines share, with the retry and dead-letter handling in one place rather than copied four times."
Handling bad data
What it screens for: The real screen, and the thing most listings fail to name.
A sentence that proves it"I added the checks that catch upstream schema changes before they reach the warehouse, after the week we silently loaded nulls into every price field."
What counts as a bonus
A cloud warehouse
What it screens for: Whether you have thought about cost, not only correctness.
A sentence that proves it"I partitioned our largest table by day and cut our warehouse bill by a third."
Streaming
What it screens for: Whether you have worked where the answer cannot wait for tomorrow.
A sentence that proves it"I built the Kafka consumer that keeps our fraud checks under two seconds end to end."
dbt or similar
What it screens for: Whether transformations are reviewed like code or edited in a console.
A sentence that proves it"I moved our transformations into dbt so every change went through review; before that they lived in a scheduled query nobody could diff."
The requirement nobody writes down
Whether you have recovered from a bad load. Everyone has pipelines that run; the question is what you did the day one of them was wrong for a week before anybody noticed.
Where candidates come up short
- Data quality work. It is the majority of the job and almost nobody writes it down.
- Cost. A pipeline that is correct and ruinous is not finished, and teams that have been burned screen for this.
- Working with the people who consume the data — the requirement hiding behind "stakeholder management".
Related roles
Data engineer: questions
- What is the one thing data engineer listings screen for without writing it down?
- Whether you have recovered from a bad load. Everyone has pipelines that run; the question is what you did the day one of them was wrong for a week before anybody noticed.
- What do candidates for data engineer roles most often fail to evidence?
- Data quality work. It is the majority of the job and almost nobody writes it down. Cost. A pipeline that is correct and ruinous is not finished, and teams that have been burned screen for this. Working with the people who consume the data — the requirement hiding behind "stakeholder management".
- How do I turn this into something a company can check?
- Tell the agent what you actually did, in sentences like the examples on this page. Each becomes a claim carrying that sentence and the date it holds for, and a company matching you against a listing sees the requirement, the claim, and your own words behind it.
Prove it, requirement by requirement.
Write sentences like the ones on this page and each becomes a claim with a date. Then match yourself against a real data engineer listing and read what is still unmet — that list is the plan.