liveby Technical Interviews

Data Science. A question about real data, answered in code and on a whiteboard.

Forty-five or sixty minutes with a Python or R editor and a whiteboard, one toggle apart. The candidate sketches the analysis, writes the code, and comes back to the diagram when the numbers surprise them. Each surface keeps its state, so moving between them costs one click and nothing else, and each gets the whole workspace rather than half of it.

Judged on whether the result would survive a skeptic

Did the candidate choose a metric that matched the decision, notice the leak, and know when a result was real? That is the judgment dimension doing its job in this round’s vocabulary, so the report calls it Evaluation judgment. It is the same dimension as Architectural Judgment in the other interview types, so a Data Science round sits on the same rubric as every other round.

Two surfaces, one toggle

  • An editor in Python 3.12 or R 4.4. R is offered here and in no other interview type, because a statistician should not be assessed in their second language
  • Switching language never destroys work: only an untouched starter buffer is replaced
  • A whiteboard with an analysis stencil set: dataset, population, train and test split, transform, feature, label, model, metric, experiment, serving, monitoring, confounder and note
  • Typed edges between them, so “data flows into” and “confounds” are different statements

What it deliberately is not

  • A drawing test. In this round the whiteboard is a thinking aid for the candidate and is not an input to the evaluation, which works from the transcript and the rubric
  • Usable on a phone. Below 680px wide the room declines to start and says so, with nothing used up, rather than producing a weak result from a cramped editor and canvas
45 or 60 minutes1 credit either way, the same as every other interview type
Python or RThe candidate chooses in the room, and can change their mind without losing what they have written
A laptop or desktopThe editor and the whiteboard are each unusable at phone width

How it is weighted

The same four dimensions in every interview type: Problem Decomposition, Evaluation Judgment, Technical Depth and Communication. The level moves the weights, and the arithmetic is code, not judgment. Read the rubric →

Send the data science round to yourself first.

The same room your candidates get.

The other interview types: Technical Screen · Coding · System Design