Roblox Intern Data Scientist Interview Experience — A Four-Part OA With Two In-Platform Games

Roblox's DS summer intern OA has 4 parts: the first two parts are both playing games on Roblox's own platform, the third part is basically behavioral multiple choice, and the fourth part is coding.

Part 1: a build-a-car game, 25 minutes total; 2 sections, 13 minutes each, and each section has two different terrains.

You need to install different parts on a car to get through both terrains, and there's a limit on how many parts the car can carry. The goal is to build as many successful cars as possible.

The parts' functions are really abstract — if you read the description once and can't understand a part's behavior, don't waste time trying to figure it out, just slap it on the car and test it, then watch the replay afterward and you'll understand exactly what the part does. Putting a part in a different position on the car also changes its effect.

There's no great strategy for this, because everyone gets assigned different parts and terrains — you just have to keep trying. Get through the first terrain before worrying about the second. Installing each part costs energy, and a car can hold at most three parts, so you need to think about how to cover as many functions as possible within those 3 parts. Some parts only work if you use two of them together — I didn't figure this out until the last two minutes, so everyone should watch out for that.

Part 2: a toy factory, 25 minutes total, one section.

A factory assembly line makes four toys, and you maximize profit over 24 hours. You need to buy raw materials, manufacture parts, combine parts, and complete the toys. Each toy has a different profit margin, so you need to calculate based on the cost of the required materials/parts and the sale price.

On the assembly line you have to coordinate each part's production quantity and production time — some parts need other parts to be combined into them, but the quantities needed and the production times of those other parts are all different. If a part overflows storage after it's produced and isn't used in time, it just gets wasted. A part only starts being produced once all the required quantities of materials for it have finished being produced.

Some parts need a lot of small components, but the max production rate of those small components is lower than what's needed, so you have to balance the coordination among a lot of parts. You can stack a buff on a part's production machine to double output or reduce the required quantity, but that also increases cost accordingly.

I personally didn't come up with any strategy at all, I just tried things randomly; but you could try getting one production line fully working first, then move on to the next one — start with the most profitable product. After playing through it I genuinely felt like my IQ was zero — I completely did not understand the logic of this game.

Part 3: behavioral multiple choice, 25 minutes total, 23 questions.

Each question gives you a scenario with 4 approaches, and you pick the best one and the worst one. Usually it's about what you should do when you've spotted an issue or you have a different idea from everyone else. It felt more like a personality test, and I have no idea what the "correct" answers even are. But my sense was that the worst option is usually whichever one isn't proactive.

Part 4: coding, 50 minutes total, four questions.

This is on CodeSignal, and you can use R or Python.

Question 1: given a numeric array, a significance level, power, and an effect size to detect, use a two-sample z-test to calculate the required sample size.

Question 2: given three arrays — period, group, outcome — calculate difference-in-differences, plus a trend validation that checks whether the pre-period difference is within a given threshold.

Question 3: use Bayes' rule to calculate a posterior probability.

Question 4: given a 2D array where each row is a feature, plus an outcome array, run a logistic regression and return the names of the top 3 features.

No formulas were provided for the coding section, so you need to memorize the various formulas beforehand.