Round 1: HR Call
Asked about "why this company," and otherwise it was just some basic logistics confirmation.
Round 2: Data Challenge
This was done on CodeSignal — about 4 questions total, using numpy, pandas, sklearn, etc. It was split into a few parts: data cleaning, feature selection, and model building. Finishing the data-processing part within the time limit (90 min) was fairly difficult; you need to be fluent with joining, filtering, and similar operations. You don't need a perfect score — there seems to be a passing line.
Round 3: Manager Phone Screen
Mostly resume-based questions. You just need to be solid on the concepts and standard interview talking points tied to each project on your resume. The questions were fairly specific, but this round was only 30 minutes, so they didn't dig too deep into any one question.
Round 4: Power Day
Case
The classic vegan burger case question — no need to elaborate on it here.
Technical
They started with environment installation and virtual environment setup, command-line stuff. Next was a Python code review — you need to quickly read the code and explain what it does; the content was related to data processing, involving numpy and pandas. After the code review, they run the tests, then ask whether the existing test cases are complete. If not, you need to add test cases, following the style of what's already in the code. Last was a code format review — another piece of Python data science code — you need to quickly read it, explain what it does, and point out the formatting and design problems in it. There were a lot of issues, so you just list them one by one.
Role Play
The classic flight-delay problem, and I chose regression. It's basically consistent with what's already been described elsewhere, so let me just add a few notes:
(1) At the end of the presentation you need to summarize the business insight — don't spend too much time on the tech here, talk instead about what the airline can actually, concretely do about it.
(2) The interviewer asks a lot of drill-down follow-up questions, so when you propose improvement actions, be ready to get pushed back on. For example, if you say "variables have high correlations so we may remove them," common pushback includes: how high a correlation counts as high? Why is high correlation bad? Why does removing it solve the problem? If I really don't want to remove it, is there another way? If I remove feature x, how do I know what its impact actually was? Basically, my suggestion is to ask yourself these kinds of questions a lot while you prepare, and build up your defenses. I did prepare some defenses, but I still ran into a few questions during the interview that I hadn't thought of.
Role Fit
Started with a short behavioral question: what kind of team do you like working in, what's the best team you remember being on, and what role did you play in it.
Then came the evergreen face-recognition question — if you were building a model to detect a robber via camera, what factors do you need to consider in the model design.
After that, some conceptual technical questions. Since I was interviewing for a specific team, they also asked some team-specific fundamentals questions that aren't broadly applicable, so I won't go into those here.
Discussion
Loading comments…