just finished my GS data scientist interview loop for the Quantitative Engineering division in NYC. sharing notes while it's fresh because i couldn't find good recent info when i was prepping.
the loop was four rounds: a recruiter screen, a take-home, and two technical rounds. no coding in the traditional leetcode sense.
take-home (48 hours): they gave me a messy dataset and asked me to clean it, run some exploratory analysis, and build a simple predictive model. they cared a lot about how i communicated uncertainty. my model wasn't fancy but i walked through every assumption and they seemed to like that.
round 1, SQL + case: this was the one i was least prepared for. the SQL was genuinely hard. think multi-table joins with conditional aggregations, window functions for rolling metrics, and one question where i had to find gaps in a time series. not just "write a query", more like "here's a business scenario, figure out what to measure." the case part was a market sizing thing. they didn't care about the exact number, they cared about the structure and whether i could sense-check my answer.
round 2, stats + ML concepts: probability questions (bayesian updating, conditional probability), some hypothesis testing, and they asked me to explain how i'd detect drift in a deployed model. also asked about a past project in depth. very much a "what did you actually do" vibe, not "tell me about a time."
no behavioral rounds at all as a DS. a little surprising.
overall: harder than i expected on the SQL side, lighter on ML depth than a typical quant fund. felt like they want people who can do rigorous analysis and communicate it clearly, not necessarily people who can tune transformers.