Two Sigma Data Scientist interview questions

31 Data Scientists have shared notes from their Two Sigma interviews, with the questions they were asked in each round. That is 50% of the 62 Two Sigma interview notes on Screna AI.

31 Data Scientist interview notes

Two Sigma·Data Scientist·Technical Phone Screen

Jul 2026
Two Sigma data science interview that went deep on statistical theory, specifically the Central Limit Theorem. More rigorous than I expected for a DS role, felt closer to a stats PhD oral exam than a typical tech interview.
  • Why do statisticians treat the sample mean as approximately normally distributed for large samples? What exactly becomes normal, what theorem justifies it, why does it hold, and when does the approximation break down?
  • Name a distribution where the sample mean never becomes approximately normal no matter how large the sample is, and explain why the theorem's conditions fail for it.
  • How fast does the normal approximation improve as sample size grows, and what property of the underlying distribution controls that rate?
  • Observations in a time series are dependent. Does a CLT-type result still hold, and what conditions are needed for it?
  • If you suspect the normal approximation is poor for your actual sample size, what would you do to build a confidence interval for the mean instead?

“This question is deceptively broad and I fumbled the opening by saying 'the data becomes normal' which is wrong and they caught it immediately.”

View Post

Two Sigma·Data Scientist·Technical Phone Screen

Jul 2026
Two Sigma data scientist interview with a coding problem that looks like a simple palindrome question until you actually read the constraints. The O(n) requirement rules out the naive increment-and-check approach entirely, which I did not fully appreciate until I was already mid-explanation.
  • Given a positive integer K, find the smallest palindrome strictly greater than K. Your solution must run in time proportional to the number of digits, not by incrementing one at a time.

“I started by thinking about mirroring the left half onto the right half, which gets you partway there.”

View Post

Two Sigma·Data Scientist·Technical Phone Screen

Jul 2026
Got a coding problem from Two Sigma for a Data Scientist role that looked like a graph question but had a nasty twist with fractional weights. The kind of problem where your first instinct is completely wrong and you have to back out of it.
  • You have a complete directed graph with n nodes (up to 15). Every pair of distinct nodes has a directed edge with a positive weight, and weights can be less than 1. The score of a simple path is the product of its edge weights. Find the maximum score over all simple paths in the graph. The optimal path might stop early since multiplying by a weight under 1 shrinks the product.

“My first instinct was to just find the longest path or something Hamiltonian, which is completely wrong once you think about weights below 1.”

View Post

Two Sigma·Data Scientist·Technical Phone Screen

Jul 2026
Two Sigma Data Scientist interview with a numerical methods coding problem. The question was more math-adjacent than pure CS, which I wasn't fully expecting. Solid problem overall, felt like it was testing whether you could translate a spec into clean code rather than just grind through an algorithm.
  • Implement a piecewise linear interpolator with extrapolation: given a set of 2D knot points in arbitrary order, sort them by x-coordinate, build a piecewise linear function connecting consecutive knots, and evaluate it at arbitrary query points. For queries outside the knot range, extrapolate using the slope of the nearest endpoint segment. The solution should run in O((n + q) log n).

“The core logic isn't that bad once you sketch it out.”

View Post

Two Sigma·Data Scientist·Technical Phone Screen

Jul 2026
Two Sigma data scientist interview that was essentially a deep OLS regression question dressed up as a coding exercise. Five sub-parts, all interconnected, and it moved fast.
  • You're given two separate dataframes, one with features (user_id, clicks, impressions) and one with a target (user_id, conversions). Inner-join them on user_id, then fit a no-intercept OLS model using both the normal equations and a numerically stable factorization like QR or SVD. Show the intermediate matrices XᵀX and Xᵀy and the final coefficient vector.
  • Compute R-squared for the no-intercept model. The correct TSS here uses sum of squared y values, not sum of squared deviations from the mean. Explain why this version of R-squared can go negative and how it compares to the standard intercept model's R-squared.
  • When is it actually appropriate to drop the intercept from a regression model? Mean-center both X and y, then refit with and without an intercept and comment on what changes with the coefficients.
  • If XᵀX is singular or near-singular due to collinearity between clicks and impressions, how do you detect that and what do you do? Derive the closed-form ridge regression estimator with regularization parameter lambda and compute it on the sample data.
  • How would you validate this model in practice? Walk through residual diagnostics and cross-validation, and explain how you'd prevent data leakage when constructing the feature and target dataframes.

“This part took longer than I expected because I kept second-guessing the join step.”

View Post

Two Sigma·Data Scientist·Technical Phone Screen

Jul 2026
Two Sigma Data Scientist interview, open-ended ML system design question about using Twitter data to predict stock prices. The whole thing was a long discussion with no right answer, which I found both refreshing and stressful. They clearly cared more about how you reason than what model you named.
  • You have a large historical stream of tweets (text, timestamps, author metadata) and aligned market data. How would you build a system to predict stock prices from Twitter data, end to end?
  • Your backtest shows a small but consistent edge. What checks would you run before trusting that the result is real?
  • Tweet activity often spikes after a price move, not before. How do you separate tweets that predict moves from tweets that are just reacting to them?
  • How would you defend your signal against coordinated manipulation like pump-and-dump schemes or bot networks in the tweet data?
  • If the signal starts decaying after deployment, how would you detect it and what would you do?

“My first instinct was to jump straight to 'train a sentiment model and regress on price' which, looking back, is exactly the wrong move.”

View Post

Two Sigma·Data Scientist·Technical Phone Screen

Jun 2026
Two Sigma data scientist interview that went deep on sorting algorithms. The whole session was basically a merge sort deep dive, which I wasn't expecting to be as involved as it turned out to be.
  • Implement merge sort on an integer array, sorting in non-decreasing order with O(N log N) time and O(N) auxiliary space. Then walk through both a recursive top-down version and an iterative bottom-up version.
  • How do you merge two sorted halves of an array in O(N) time using a single auxiliary buffer?
  • Is merge sort stable? How does that compare to quicksort, and why does stability matter?
  • When would you prefer merge sort over quicksort or heapsort? Walk through the tradeoffs.

“The recursive version came out fine, I've written it enough times.”

View Post

Two Sigma·Data Scientist·Technical Phone Screen

Jun 2026
Two Sigma DS interview centered on a bike-share demand forecasting case. The whole thing was one long open-ended problem and they kept pushing with follow-ups until you ran out of things to say. Harder than I expected for a single question.
  • Walk through end-to-end how you'd build a model to predict bike pickup demand at a single Citi Bike station, including problem framing, features, model choice, and evaluation.
  • How would you estimate true demand for hours when the station had zero bikes available, and how would you validate that correction?
  • How would you scale this from one station to the full network? One model per station versus a single global model with station-level features or embeddings, and what are the trade-offs?
  • A nearby subway line shuts down for a month and demand spikes. How does your system detect that the model is stale, and how do you adapt?
  • If a stockout costs five times as much as an idle bike, how does that change your modeling target and evaluation metric?

“This is the kind of question where you can talk for 20 minutes and still feel like you only scratched the surface.”

View Post

Two Sigma·Data Scientist·Technical Phone Screen

Jun 2026
Two Sigma data scientist interview with a probability/combinatorics problem that looked like a graph question but was really just linearity of expectation dressed up in fancy clothing. Clean problem statement, no behavioral stuff from what I could tell.
  • In a complete graph on n vertices where every edge weight is drawn independently from the same continuous distribution, define a 'good pair' as two vertices that are each other's maximum-weight neighbor. Derive a closed-form expression for the expected number of good pairs.

“I spent the first few minutes thinking about this combinatorially in a way that was going nowhere.”

View Post

Two Sigma·Data Scientist·Technical Phone Screen

Jun 2026
Two Sigma data scientist interview that went deep on algorithmic problem solving. The matrix question they threw at me was harder than I expected for the role, felt more like a software engineering screen than anything data-related.
  • Given a 2D binary matrix of 0s and 1s, find the largest contiguous rectangular submatrix containing only 1s and return its area.

“I knew the histogram trick going in but still fumbled explaining it under pressure.”

View Post