Point72 Interview Questions — Page 2

Point72·Data Scientist·Technical Phone Screen

Apr 2026

Technical screen for a Data Scientist role at Point72, heavily focused on decision trees. The questions went deep fast, like they assumed you'd already built one in production and wanted to stress-test every assumption you made along the way.

How does a CART decision tree choose splits differently for classification versus regression? Walk through the exact formulas for Gini impurity, entropy, and MSE, and explain how surrogate splits handle missing feature values.
Walk through a defensible procedure for choosing max_depth and min_samples_split, including a cross-validation plan, cost-complexity pruning via the alpha path, and which metric you'd optimize under severe class imbalance and why.
What are at least three diagnostics you'd use to detect overfitting in a decision tree, and what specific patterns in those diagnostics would flag a problem?
With 500k rows and around 300 features including high-cardinality categoricals and sparse indicators, how would you preprocess and train a single decision tree? Cover encoding choices, rare category handling, feature binning, any monotonic constraints, and give rough hyperparameter ranges and expected training time.
Under what data or target conditions would a random forest or gradient-boosted model outperform a single decision tree? Name at least three, and discuss the trade-offs around variance, interpretability, latency, calibration, and how you'd compare the models fairly.

“I knew Gini and entropy cold but surrogate splits caught me mid-sentence.”

View Post

Point72·Machine Learning Engineer·Technical Phone Screen

Apr 2026

Point72 ML engineer interview with a mix of probability and linear algebra. The questions were more math-heavy than I expected for an MLE role, felt like a quant screen in disguise.

Given two independent standard normal random variables X and Y, what is P[X > 3Y]?
In ordinary linear regression with a design matrix, response vector, and coefficient vector, how do you estimate the coefficients?
When the number of observations is much larger than the number of features, how would you compute the regression coefficients efficiently and exactly?

“Blanked for a second on this.”

View Post

Point72·Data Scientist·Technical Phone Screen

Apr 2026

Point72 data scientist round with a combinatorics/greedy algorithm problem that looks deceptively clean until you hit the edge cases. The problem had real depth to it and required thinking carefully about optimality proofs, not just a working solution.

Given item counts by color, find the maximum number of outfits you can form where each outfit uses exactly 3 items of distinct colors. Return the count and construct the first 10 valid outfits. Aim for O(C log C) time and O(C) space.

“I got the greedy idea pretty fast, grab the three largest remaining counts each round using a max-heap, but then they asked me to prove it's optimal and I fumbled.”

View Post

Point72·Data Scientist·Technical Phone Screen

Apr 2026

Point72 Data Scientist interview with a SQL question centered on department-level aggregation. Pretty standard stuff but the LEFT JOIN detail matters more than it looks.

Write a SQL query that shows each department name alongside its student count, making sure departments with zero students still appear in the results.

“The zero-count departments are the whole point.”

View Post

Point72·Software Engineer·Onsite - System Design / Architecture

Apr 2026

System design interview at Point72 for a data engineering role. The whole thing was focused on one big open-ended design problem around pipeline orchestration, data quality, and observability. Pretty deep technically, and the streaming failure handling question at the end had some real nuance to it.

How would you design a production orchestration system for batch and streaming data pipelines that handles upstream and downstream dependencies?
What data quality checks would you build into your pipelines? Walk through how you'd implement things like threshold checks, schema validation, null checks, primary key checks, and upstream dependency checks.
What would a data observability framework look like in practice?
If 2% of records in a streaming pipeline fail validation, do you stop the entire pipeline or keep it running? How do you decide, and what do you do with the failed records?

“This is where I started rambling a bit.”

View Post

Point72·Data Scientist·Take-home Assignment

May 2026

Point72 data scientist take-home with three pretty brutal coding tasks. No fluff, no behavioral stuff, just pure problem-solving under time pressure. The problems were harder than I expected for a DS role.

Given a list of date strings in mixed and potentially ambiguous formats, normalize each one to UTC ISO 8601. Handle timezone conversions, two-digit year mapping, leap year validation, and compute the business-day difference between any two normalized dates, excluding weekends and a fixed holiday set. Return an error for invalid inputs.
Given target quantities for up to 6 item types, unit prices, and a list of special bundles each with their own price, find the minimum cost to exactly meet the target quantities using any combination of bundles and individual unit purchases. Return the cost and the actual combination used, or -1 if it's impossible. Bundles cannot oversupply.
Implement a function that classifies the spatial relationship between two circles, returning one of seven possible states from 'separate' to 'concentric equal'. Use squared distances to avoid precision loss, handle near-tangency within a small epsilon, and return intersection point coordinates when applicable.

“This one wrecked me more than I want to admit.”

View Post

Point72·Software Engineer·Onsite - Behavioral / Leadership

May 2026

Behavioral round for a Data Engineer role at Point72. Just one question but it was a meaty one, the kind where you can't really bluff your way through with a vague answer.

Describe a situation where you had to build a solution that worked for multiple stakeholders with competing priorities. Walk through your role, the trade-offs you faced, how you made decisions, and what happened in the end.

“This is the type of question where picking the right story matters more than how you tell it.”

View Post

Point72·Machine Learning Engineer·Technical Phone Screen

May 2026

Point72 ML Engineer interview with a prompt engineering question that was more open-ended than I expected. The whole thing felt less like a coding screen and more like a product-adjacent design exercise, which threw me a bit.

You are coaching an Olympic champion. The athlete receives a high volume of news articles daily. Design a prompt for a large language model that filters incoming articles, showing only relevant and actionable content to the athlete and coaching team. The prompt should define the model's role, specify what to include and exclude, require structured output (decision, relevance score, category, summary, reason), and include safeguards against speculation and sensationalism. Assume each article has a headline, source, date, and body text.

“This was the whole interview basically.”

View Post

Point72·Machine Learning Engineer·Technical Phone Screen

May 2026

Interviewed for an ML Engineer role at Point72 and got hit with some Transformer theory questions. Not the vibe I was expecting from a quant shop but they clearly care about fundamentals.

What are the key differences between a Transformer encoder and decoder, specifically around attention mechanisms and the use of a causal mask?
Why does the same input prompt produce different outputs from an autoregressive language model on different runs, and under what conditions would the output be deterministic?

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

View Post

Point72·Software Engineer·Online Assessment (OA)

May 2026

SQL-heavy online assessment for a Data Engineer role at Point72, one problem but it had enough edge cases to keep me busy for a while. The problem was framed as fraud detection on a crypto exchange and the output formatting requirements were brutal.

Given a transactions table on a cryptocurrency exchange, write a PostgreSQL query to detect potentially fraudulent activity: find all sender-recipient pairs with 2 or more transactions, then produce one row per sender showing their wallet address, a formatted time range from first to last transaction, and a concatenated string of transaction amounts sorted descending. For senders with 3 or fewer transactions just list them all with a sum; for more than 3, show the top 3 and collapse the rest into a '+ .. X more = Sum' suffix. Order final output by total amount descending.

“The core logic, CTEs for filtering pairs then joining back to detail rows, wasn't the hard part.”

View Post