Just finished the Replit DS interview process (applied in late 2025, finished loop in Feb 2026). Sharing because there's almost nothing out there specific to the DS track.
The loop is 3-4 rounds. Here's what I got:
SQL screen: First technical touchpoint. Two SQL questions, about 45 minutes. Not hard SQL. Window functions, aggregation, a subquery. The second one involved computing some kind of retention metric, which makes sense given the product. If you can write clean SQL and explain your reasoning, you're fine.
Statistics / experimentation round: This is where it got interesting. They asked about A/B testing setup for a product change. Specifically: how would you design an experiment to measure whether a new collaboration feature increases session depth? We talked about randomization unit (user vs session vs classroom), minimum detectable effect, and what to do about network effects (users sharing repls with each other makes clean treatment/control split harder than usual). That network effects piece is not a standard DS interview question and it took me a second to think through it properly.
Product / case round: Given a dataset description (user events: opens, edits, runs, shares, invites), tell me what you'd look at first if retention was declining. This is a diagnosis exercise more than a SQL exercise. They wanted structured thinking: segment first, form hypotheses, test them in sequence.
Behavioral: Standard questions about stakeholder communication, presenting analysis to a non-technical audience, a time your analysis led to a decision that turned out wrong.
Comp discussed was $145-160k base for a mid-senior DS role, fully remote. The experimentation round is the hardest part. If you haven't thought about network effects in A/B tests, read up before you go in.