Pinterest Interview Questions — Page 6

Pinterest·Machine Learning Engineer·Onsite - System Design / Architecture

May 2026

Pinterest ML engineer interview focused on ad system design, specifically around building embeddings for CTR optimization. The technical depth was real and they clearly expected familiarity with graph-based recommendation architectures.

Design a system to improve click-through rate for ads on a social media platform.
How would you build embeddings for users and content in this ad system? Walk through the architecture and training approach.
Where does your training data come from for the CTR model?

“Big open-ended question and I went straight into feature engineering and model selection, which felt fine at the time.”

View Post

Pinterest·Machine Learning Engineer·Technical Phone Screen

May 2026

Pinterest ML Engineer interview with a tricky string manipulation problem that took me a while to fully wrap my head around. The math behind it is deceptively subtle and I'm not sure I nailed the explanation.

You're given a string of the form 'A+B' where A and B are digit strings with no zeros. Insert exactly one pair of parentheses that must enclose the plus sign, such that digits outside the parentheses are treated as multiplicative factors. Find the placement that minimizes the resulting value and return the full expression string.

“The problem looks like a parsing puzzle but it's really an optimization problem once you see the structure.”

View Post

Pinterest·Software Engineer·Technical Phone Screen

May 2026

Pinterest SWE interview with a coding problem that boils down to heap usage. Pretty focused session, one problem with a complexity discussion tacked on at the end.

You have a list of pins, each with a height value. Distribute them across a fixed number of columns by always appending the next pin to the shortest column. Implement this and analyze the time and space complexity.

“Once I saw 'always pick the shortest column' I knew it was a min heap and coded it up pretty fast.”

View Post

Pinterest·Machine Learning Engineer·Onsite - System Design / Architecture

May 2026

Pinterest ML Engineer interview focused almost entirely on designing a home feed ranking system end to end. It was a long, dense session and I felt like I was playing catch-up the whole time trying to cover retrieval, ranking, and re-ranking without running out of time.

Design a personalized ranking system for the Pinterest home feed, covering candidate generation, ranking pipeline, and re-ranking.
What features would you use for the user, pin, and context components of your ranking model?
How would you handle the cold start problem for new pins with no engagement history?
How would you train a multi-objective ranking model and handle conflicting signals like engagement rate versus save rate?
How would you evaluate the ranking system both offline and online?
What are the key latency versus model complexity trade-offs in a system like this, and how would you meet a sub-200ms p99 latency target?

“This ate up the entire session.”

View Post

Pinterest·Data Scientist·Technical Phone Screen

May 2026

Pinterest data scientist interview with a meaty coding problem centered on set similarity at scale. The question had multiple follow-up layers and the constraints were not trivial to satisfy.

Given a large collection of named lists (up to 200k lists, 5M total items), find the pair of lists with the greatest item overlap. Return the overlap count and Jaccard similarity, with tie-breaking by higher Jaccard then lexicographic order of list names. An O(N^2) all-pairs brute force is explicitly off the table. Describe your algorithm, analyze complexity, and implement it in Python.
Extend the solution to return the top-k most similar pairs instead of just the single best pair.
How would you adapt the solution if the input arrives as a stream of (list_name, item) update events where lists grow over time?

“My first instinct was brute force and I almost said it out loud before catching myself.”

View Post

Pinterest·Machine Learning Engineer·Technical Phone Screen

May 2026

Pinterest ML Engineer oral interview, pretty much a fundamentals check on regularization and overfitting. Nothing crazy algorithmic, just conceptual depth. Felt like they wanted to see if you actually understood the math behind the stuff you use every day.

What is the difference between L1 and L2 regularization, and when would you choose one over the other?
How do you detect overfitting in a model, and what are your options for addressing it?
How does regularization relate to the bias-variance tradeoff?

“This went okay but I fumbled the prior derivation part.”

View Post

Pinterest·Data Scientist·Technical Phone Screen

May 2026

Pinterest DS interview focused almost entirely on metrics and experimentation around home feed freshness. Five questions, all technical, no behavioral fluff. The A/B test interpretation was the meatiest part and the one I spent the most time second-guessing myself on.

How would you define a practical metric for content freshness in a home feed? What counts as fresh, what user action or exposure would you measure, and what time window would you pick and why?
What are the weaknesses of the freshness metric you defined? Think about how heavy vs. light users, creators with different posting cadences, and different content categories might distort it.
Pinterest is thinking about showing more video pins in the home feed to boost engagement. What would your null hypothesis be, and can you state at least two alternative hypotheses covering both positive and negative outcomes?
Given these A/B test results from a 14-day user-level experiment, interpret each metric's p-value and say whether you'd recommend launching the change.
Walk through how you'd calculate sample size for an A/B test. How do baseline variance, minimum detectable effect, significance level, power, traffic allocation, trigger rate, variance reduction techniques, and testing multiple metrics all factor in?

“I went with something like share of impressions where the pin was created within the past 7 days, measured at the session level.”

View Post

Pinterest·Software Engineer·Technical Phone Screen

May 2026

Pinterest SWE interview with a string/search problem that looks easy until you think about scale. The follow-up on complexity is where it gets interesting.

Given a sorted list of words and a list of query prefixes, return for each prefix the index of the first word in the list that starts with that prefix, or -1 if none exists. Then discuss the time complexity of your solution.

“My first instinct was just linear scan and I almost coded it up before catching myself.”

View Post

Pinterest·Software Engineer·Onsite - System Design / Architecture

May 2026

Pinterest system design round focused entirely on building out a recommendation system for the home feed. Pretty comprehensive scope, they wanted everything from metrics to serving architecture, which made it feel less like a single question and more like a full product review.

Design a recommendation system for Pinterest's home feed, covering goals, metrics, data sources, candidate generation, ranking, and serving infrastructure.
How would you handle diversity, freshness, and novelty controls in the feed ranking without tanking engagement metrics?
What defenses would you build against abuse and spam in a content recommendation pipeline?
How would you design the offline and online evaluation strategy, including A/B testing, for this recommendation system?
Walk through the low-latency serving architecture for the recommendation system, including the feature store and model serving layer.

“This is a beast of a question.”

View Post

Pinterest·Data Scientist·Technical Phone Screen

May 2026

Pinterest DS interview that was basically a seven-part interrogation disguised as one question. They handed me my own resume and said 'pick your hardest ML project' and then just... kept going. It was a lot.

For your most complex ML project, define the business objective, target variable, key constraints, and explain which primary success metric you chose and why (for example, PR-AUC vs ROC-AUC vs cost-weighted error).
Describe the data pipeline for that project: sources, labeling strategy, train/val/test splits. If it's time-series data, how did you do the split and what specific leakage risks did you check for?
Walk through your model selection process: what candidates did you consider, which hyperparameters did you tune, and can you show an ablation plan that isolates the value of two specific feature groups?
How did you handle class imbalance in that project? Walk through your resampling or weighting approach and threshold selection. Then compute F1, TP, FP, FN counts, and total cost for two model configurations on a 10,000-example validation set with 8% positives, where FP costs 1 and FN costs 5.
How would you monitor this model in production? Propose specific metrics for calibration and feature drift, define alert thresholds, and describe your retraining trigger logic and schema change handling.
Design an A/B test to validate this model online. Include guardrail metrics, how you'd estimate sample size and test duration, and what your rollback plan looks like if a long-tail user segment regresses.
Name two realistic failure modes for this model and explain how you'd debug each using offline error analysis and online traffic slices.

“This part sounds easy until you realize they want the 'why' on the metric choice, not just the name.”

View Post