GitLab · Primly Community

GitLab data scientist interview (SQL + case + stats): what the loop looks like

ds_dmitri · 4 replies

went through the GitLab DS loop for a mid-senior DS role on the product analytics team earlier this year. not a lot of specific info out there so here's what i found.

the structure: recruiter screen (30 min) hiring manager (45 min, background + how you think about problems) SQL + technical interview (60 min, live) product case interview (60 min) statistics / ML concepts interview (45 min) behavioral / values (45 min)

five substantive rounds is a lot. budget for a full day if you're doing it back-to-back.

SQL round: this was harder than average. not 'SELECT * FROM users WHERE date > X' stuff. we did: a multi-table join across user events, pipeline runs, and project metadata to answer a business question about activation drop-off. then a follow-up where i had to rewrite a slow query (it had an N+1 pattern baked in). then a window function question: 'for each user, find the time between their first pipeline run and their first successful pipeline run.' that last one is where i nearly fumbled.

the SQL environment was just a shared doc. no live query runner. so you're writing it out and explaining, not running it.

product case: prompt: 'GitLab's free-to-paid conversion rate dropped 8% last quarter. walk me through how you'd diagnose it.' totally open ended. they want funnel thinking, cohort thinking, and hypothesis generation. i ran through it for 25 minutes and then they started asking 'what if i told you the drop happened only on mobile?' type steering questions. classic case interview cadence.

stats / ML interview: no coding. pure concepts. questions: explain precision vs recall in the context of a spam classifier. explain how you'd detect if an A/B test had a novelty effect. 'your model shows 92% accuracy, why might that number be misleading?' (imbalanced classes, obviously, but they want you to reason through it). also: bias-variance tradeoff, when would you choose a simpler model over a complex one.

overall difficulty: harder than i expected, especially the SQL. prep with window functions and multi-table aggregations. the case and stats rounds were well within normal DS interview difficulty.

comp: my offer was around $165k base for mid-senior level, US remote, in 2026. RSUs on top. not FAANG numbers but fully remote and the work is interesting if you care about devtools metrics.

4 replies

analyst_ana

the 'write SQL in a doc without running it' format is so stressful. i always make small syntax errors that a linter would catch instantly. any tips for getting comfortable with that format?

ds_dmitri

practice writing SQL in a text editor with no autocomplete for a week before. google sheets works if you want something bland enough. the main thing is getting used to saying 'i'd double check this syntax but the logic is...' and keeping moving. interviewers generally know what you mean if the logic is right.

de_derek

that N+1 query cleanup question is a great real-world test. way better than asking someone to implement a red-black tree. it's the kind of thing that actually shows up in production GitLab work (their data infra is Ruby/Rails heavy in some places and N+1s are a classic pain point).

ml_mike

the stats round sounds pretty standard. the novelty effect question is interesting though, i haven't seen that one as a standard prompt. sounds like they've been burned by it before. always a sign a company is actually sophisticated about experimentation vs just claiming to be.