Voleon interview notes, page 4 of 4

Small

Most of these Voleon write-ups come from Software Engineers (24), then Data Scientists (10) and Machine Learning Engineers (2). The most reported rounds are the technical phone screen (28), the system design round (3), the multi-round onsite (2) and the take-home assignment (2).

37 interview notes · updated Jul 2026

Voleon·Data Scientist·Technical Phone Screen

Jun 2026

Voleon Group data scientist interview, one long technical screen that was basically a pandas gauntlet. Five interconnected problems on a single dataset, and they clearly care a lot about production-level thinking, not just getting the right answer.

Given two DataFrames (events and users), clean and normalize all timestamps to be timezone-aware, deduplicate events defined as the same (user_id, event, session_id) within a 5-minute window keeping the earliest timestamp, drop events for users not present in the users table, and explain how you'd avoid chained assignment throughout.
Define an 'active day' as any local calendar day (in each user's own timezone) with at least one non-signup event. Compute daily active users per local date for a specific date range, excluding bots and excluding device IDs that appear across 20 or more distinct users (shared devices).
Compute 7-day retention for signup cohorts over a specific week. A user counts as retained if they have any non-signup event on exactly the 7th day after signup, measured in their local timezone. Return cohort size, retained count, retention rate, and a Wilson 95% confidence interval.
Parse revenue strings that may include currency symbols and locale-specific formatting (e.g. commas as decimal separators), convert to USD using a given FX rate mapping, impute missing revenue for purchase rows using median revenue over the prior 30 days within country-plan groups, then compute ARPU for the last 7 days for non-bot users.
For a hypothetical dataset of 100 million events, provide vectorized pandas code sketches (no Python loops), discuss the expected memory footprint and computational complexity, and outline a chunked processing strategy to keep peak RAM under 8 GB using techniques like categorical dtypes, dtype maps on read, sorted merges, and groupby with observed=True.

“The chained assignment part is what got me.”

View Post

Voleon·Data Scientist·Technical Phone Screen

Jun 2026

Voleon Group DS tech screen, focused heavily on statistical inference. The questions were more conceptual than I expected, less coding and more 'do you actually understand what the output means.'

You see a p-value reported as exactly 0.000 for a regression coefficient. Is that a real value? Why does software print it that way, and how should you actually communicate statistical significance in that situation?
Walk through how you'd use bootstrapping to estimate the sampling distribution of a regression coefficient. How does the bootstrap expectation relate to the true population parameter, and what does that tell you about bias?

“I knew this was about rounding but fumbled the explanation a bit.”

View Post

Voleon·Data Scientist·Technical Phone Screen

Jun 2026

Voleon DS tech round, one question, very applied. They gave you a small financial time-series table and asked you to write the full preprocessing pipeline in pandas before fitting a linear regression. Not conceptual at all, just get it done.

Given a financial time-series dataset with date, two feature columns, and a target column, write Python/Pandas code to drop nulls, winsorize extreme values at the 1st and 99th percentiles, standardize the predictor columns, and produce an X and y pair ready for linear regression.

“The winsorizing part is where I fumbled a bit.”

View Post

Voleon·Software Engineer·Technical Phone Screen

Jun 2026

Voleon SWE interview had at least one coding round that went straight into a circular game simulation variant. Not a brutal interview but the twist on a known problem meant you couldn't just recite the standard solution.

Given n friends sitting in a circle, simulate an elimination game where the count-to-eliminate changes each round according to a given array. Return the last remaining friend.

“I recognized the base problem pretty fast but the per-round k array threw me for a second.”

View Post

Voleon·Software Engineer·Technical Phone Screen

Jul 2026

Voleon threw a recursive string parsing problem at me that looked deceptively clean on the surface. The constraint to hit O(n) was the real test.

A 'good string' is defined recursively: '0' is a good string, and if A and B are both good strings, then '1' + A + B is also a good string. Given a binary string, determine whether it is a good string. Your solution must run in O(n) time.

“My first instinct was to write a recursive parser and call it a day, which technically works but I fumbled explaining the O(n) guarantee because I kept second-guessing whether the recursion would revisit characters.”

View Post

Voleon·Data Scientist·Technical Phone Screen

Jul 2026

Voleon Group DS technical round focused on linear regression applied to financial data, specifically around fitting OLS models and interpreting results in an economic context. Pretty rigorous for a single question but they really wanted depth on the statistics side.

Fit an OLS regression model with a target variable as the dependent variable and all remaining columns as predictors. Walk through each coefficient, explain what they mean economically, and describe how you'd test the marginal contribution of each feature sequentially.

“This is where I spent most of the interview.”

View Post

Voleon·Data Scientist·Technical Phone Screen

Jul 2026

Voleon data scientist interview with three pretty heavy technical questions back to back. The probability and linear algebra stuff felt like a grad school exam more than a job screen. Not sure how I did.

What are the defining properties of a Markov transition matrix, and what can you say about its eigenvalues, stationary distributions, and the convergence of its powers as k goes to infinity?
If you rotate your two-dimensional feature space using an orthonormal rotation matrix and refit OLS with an intercept, how do the estimated slope coefficients change, are the fitted values preserved, and under what distributional conditions on the features is the sampling distribution of the slope estimator invariant to rotation in second moments?
Players A and B flip fair coins simultaneously each round. A stops at the first round where they've seen two consecutive heads, B stops at the first round where they flip a tail. What is the probability that A finishes strictly before B?

“I knew the basics: non-negative entries, rows sum to one, eigenvalue of 1 always exists.”

View Post