What we test for
Typical work
- A DAG that fails silently and is noticed when the numbers are already wrong
- Backfilling a year of history without duplicating a row
- Splitting one hour-long task into steps that can be retried on their own
- Moving orchestration off cron jobs on a server nobody wants to touch
The coding test
A nightly DAG failed four days ago and nobody noticed until a report was wrong. Make the failure loud, then make the backfill safe to run twice. Then split the one task that takes an hour into something you can retry from the middle. Scored on idempotency.
Five interview questions you can reuse
We use questions like these in our soft-skills and technical rounds. Steal them for your own interviews.
What makes a task idempotent, and why is it not optional?
A strong answer covers: Running it twice for the same logical date leaves the same result: write to a partition you overwrite rather than append to, use a merge keyed on something natural, or stage and swap. It is not optional because retries, backfills and a cleared task are all normal operations, and a pipeline that cannot be re-run is a pipeline nobody dares to fix.
A DAG has been failing for four days and nobody noticed. What was missing?
A strong answer covers: An alert that reaches a person who can act, and an SLA so lateness is a failure rather than silence. Email on failure in a mailbox nobody reads is the usual answer. The second half is a freshness check at the consumption end, because the pipeline can succeed and still deliver nothing.
How do you size a task?
A strong answer covers: Small enough to retry without redoing an hour of work, large enough that the scheduler is not the bottleneck. The practical test is whether a failure halfway through costs you one step or the whole night. Splitting by partition usually gives you both parallelism and a retry boundary at the same time.
When is a sensor the wrong tool?
A strong answer covers: When it sits there occupying a worker slot for hours waiting for a file, which is how a pool starves and everything else is late. Deferrable operators, or an event that triggers the DAG, cost nothing while they wait. The question to ask is what happens when the upstream is four hours late, because it will be.
How should two DAGs that depend on each other be connected?
A strong answer covers: By data, not by schedule. A dataset or an explicit sensor on the upstream run, so a late upstream delays the downstream instead of feeding it yesterday's table. Two cron schedules twenty minutes apart is a bet that nothing will ever be slow, and the day it is, nothing fails and the numbers are quietly wrong.
One invoice a month.
The shortlist is free. Pick the contract that suits you: monthly, six months or twelve.
What every seat includes
- Contracts, payroll and compliance handled
- Vetting: coding test, soft skills, English and a level check
- Free replacements: unlimited on 12+ months, 2 on 6 months, 1 on monthly
- A direct line to our CEO
Terms in plain words
- Monthly, six months or twelve, your choice
- Your rate is locked for the whole contract on 6 and 12+ months
- Monthly ends with 30 days’ notice
- Want to hire the developer full-time? The buyout is set in the contract from day one
Questions tech leads ask us
Is this the same pool as your data engineers?
Yes, scored on orchestration rather than on modelling. If the brief is really about the warehouse, the data engineering or dbt test is the better signal.
Airflow, Dagster or Prefect?
We staff Airflow because that is what most teams run. A senior will tell you what the others do better and will not pretend a migration is free.
Skip the sales rep. Talk to the CEO.
Tell us what you're building, your stack and when you need people. Three vetted profiles per role within 48 hours of the call, and a straight answer on whether we're the right fit.