Roblox Data Scientist Interview Questions

Roblox·Data Scientist·Onsite - System Design / Architecture

Jul 2026

Roblox data scientist interview with a deep ML system design question. One question, very technical, the kind where you can tell they actually want to see if you've built this stuff before and not just read about it.

Design a full end-to-end ML pipeline to predict whether a machine will experience a fault within the next 24 hours, evaluated at each hour. Your design needs to cover leakage prevention with a feature store strategy, time-based cross-validation with explicit cutoffs, handling ~1% class imbalance with cost-sensitive thresholding (FN=$10,000, FP=$500), probability calibration and feature importance with SHAP caveats, and robustness to sensor missingness, outliers, and distribution drift.

“This was a lot.”

View Post

Roblox·Data Scientist·Technical Phone Screen

Jul 2026

Roblox data scientist interview that went deep into auction theory, specifically GSP vs VCG mechanics with quality scores. The questions were more rigorous than I expected for a DS role, felt closer to a research economist interview at points.

Given a two-slot auction with position CTRs, three advertisers with bids and quality scores, and a reserve price, walk through the GSP allocation and per-click pricing. Then compute expected revenue and total welfare.
Using the same auction setup, compute each winner's per-click price under VCG with quality scores and a reserve price. How does platform revenue compare to GSP?
Under GSP, assuming the other advertisers bid truthfully, is there a profitable bid shading strategy for advertiser B? Find a shading range that keeps B in the same slot but lowers their price.
How does budget pacing (a multiplicative bid dampening factor between 0 and 1) interact with quality-adjusted ranking in an auction? Between GSP and VCG, which mechanism holds up better for truthful reporting when advertisers are budget-constrained?

“This took me longer than it should have.”

View Post

Roblox·Data Scientist·Technical Phone Screen

Jul 2026

SQL-heavy technical screen for a DS role at Roblox. Two questions, both around a social graph schema with users, posts, engagement events, and follows. Nothing behavioral, just pure query writing the whole time.

Given a social product schema with users, posts, engagement events, and follows tables, write a SQL query that computes each user's influence score, defined as the total number of engagement events their posts received plus their total follower count.
Using the same schema, write a SQL query that computes monthly follower growth per user, including new followers gained each month, the previous month's new follower count via LAG, a month-over-month growth rate with divide-by-zero handling, and a cumulative running total of followers.

“My first instinct was to do both aggregations in one big join and i immediately tangled myself up with duplicate counts.”

View Post

Roblox·Data Scientist·Technical Phone Screen

Jul 2026

Roblox data scientist interview with a pretty gnarly SQL question that tests your ability to handle deduplication logic, window functions, and multi-step aggregation all at once. One question, but it's a lot of question.

Write a SQL query to compute, for each campaign over a 7-day window, total impressions, total charged clicks (deduplicating multiple clicks on the same ad request to only count the earliest), CTR, RPM, and standard deviation of daily RPM. Filter to campaigns with at least 10,000 impressions and order by RPM descending then stddev ascending.

“The dedup rule is what makes this hard.”

View Post

Roblox·Data Scientist·Technical Phone Screen

Jul 2026

Got a technical screen for a Data Scientist role at Roblox that was pretty much one big coding/stats problem. The whole thing centered on implementing a difference-in-differences estimator with a pre-trend check, which I wasn't expecting to write from scratch in an interview setting.

You're given panel data as three arrays (period, group, outcome). Implement a difference-in-differences estimator using the latest pre-period and earliest post-period, then validate parallel pre-trends by checking whether the max gap in treatment-control differences across all pre-periods falls within a given threshold.

“This wrecked me a little at first.”

View Post

Roblox·Data Scientist·Technical Phone Screen

Jul 2026

Roblox data scientist technical screen, basically one big coding problem about implementing p-values from scratch in Python. Pretty brutal if you haven't touched stats internals in a while.

Implement a Python function that computes the p-value for one-sided and two-sided hypothesis tests, supporting both z and t distributions, without relying on external libraries for the z-distribution CDF.
How do you handle edge cases like NaN or infinite inputs, degrees of freedom less than 1, and extreme test statistic values in a statistical computation function?
Walk through how you would verify the correctness of your implementation using unit tests, including specific known values and monotonicity checks.

“This was the whole interview.”

View Post

Roblox·Data Scientist·Technical Phone Screen

Jun 2026

Roblox data scientist interview threw a pretty dense SQL question at me that combined fraud logic, age calculations, and chargeback lookups all in one query. No window functions allowed, which was the annoying constraint. Felt okay about the structure but definitely second-guessed some of the edge case handling afterward.

Using only CASE WHEN and JOINs (no window functions), write a single SQL query that returns, for each user with at least one payment in the last 7 days, their user_id, total payment count in that window, count of suspect payments in that window, and a Y/N flag for whether they have any chargeback on payments from the last 60 days. A payment is suspect if the user was under 18 at the time of payment, the user's country doesn't match the payment's ip_country, or the payment happened within 24 hours of account creation and exceeded $10. The chargeback flag covers any chargeback timestamped on or before today linked to payments in the 60-day window.

“This one took me a minute to even parse.”

View Post

Roblox·Data Scientist·Technical Phone Screen

Jun 2026

Roblox data scientist interview with a coding/technical problem that looked like a stats exercise but turned into a full ML pipeline question. Pretty involved for a single prompt.

You're given a labeled binary classification dataset with numeric features. Walk through how you'd z-score normalize the features, fit a logistic regression, then rank and return the top 3 features by coefficient magnitude. Also explain any modeling choices you'd need to make explicit, like how you handle regularization, the intercept, or zero-variance columns.

“More involved than I expected.”

View Post

Roblox·Data Scientist·Technical Phone Screen

Jun 2026

Roblox data scientist interview with a stats-heavy coding question around causal inference. The problem was well-defined but had enough edge cases to trip you up if you weren't thinking carefully about the data.

Given three parallel arrays representing period (pre/post), treatment group, and outcome values, implement a difference-in-differences estimator that computes the treatment effect across the four group-period cells.
Extend the DID implementation to flag potential pre-trend violations: compute the pre-period difference between treated and control groups and return a boolean indicating whether it exceeds a given threshold.

“The formula itself isn't hard once you see it written out.”

View Post

Roblox·Data Scientist·Technical Phone Screen

Jun 2026

Roblox DS interview with a single but very dense question that basically asks you to narrate an entire project lifecycle, defend your metric decisions under pressure, and handle a stakeholder conflict, all in one breath. Felt more like a written case study than a conversation.

Walk through an ads or growth analytics project you owned end-to-end under vague requirements and a tight deadline. Cover scope, stakeholders, success criteria you defined yourself, key technical decisions and trade-offs, how you resolved a conflict with Product or Sales, and quantify the impact. Then: if your primary guardrail metric dropped 0.7 percentage points while revenue went up 8%, what do you do and how do you communicate it?

“This is basically three questions stapled together and they want all of it.”

View Post