Goldman Sachs Data Scientist Interview Guide

Our candidates consistently report that Goldman Sachs is far more interested in the derivations behind your modeling choices than in surface-level familiarity with algorithms. Across multiple 2026 experiences, the same themes keep surfacing: loss functions, logistic regression, tree-based methods, and the tradeoffs between boosting and bagging. What stands out is the level of rigor around why a method fits a problem — one candidate specifically noted being pushed to derive the intuition behind a loss function from scratch, which is a strong signal that this team wants people who understand the math beneath the model, not just the model names.

A second pattern we've seen is that Goldman Sachs screens for composure and precision under pressure, not just raw knowledge. One candidate described navigating a probability tree and Buffon's needle problem while being expected to narrate their reasoning step by step, while another faced rapid-fire regression questions that rewarded clean, confident answers over long pauses. That combination — deep ML theory in one room, applied probability and statistics in another — tells us the Superday is deliberately designed to stress-test both dimensions. The non-obvious make-or-break is almost never a missing buzzword; it's a small conceptual slip when justifying a modeling choice or explaining an objective function under time pressure. Multiple candidates who cleared early rounds noted that a single rushed or imprecise answer in the final stage was enough to change the outcome, which makes deliberate, out-loud explanation practice especially valuable here.