Pinterest Interview Questions — Page 18

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

Jun 2026

Pinterest ML engineer interview focused entirely on a notification system design problem. It was a deep, multi-part question that covered basically the full ML lifecycle, from objective setting to serving infrastructure. Pretty intense for a single session.

Design an ML system that decides which push or email notifications to send to each Pinterest user, when to send them, and how often, with the goal of maximizing long-term engagement without hurting retention.

“This thing had like six sub-problems inside it and I didn't pace myself well.”

View Post

Pinterest·Product Manager·Onsite - Product Sense / Strategy

Jun 2026

Pinterest PM interview, one question about metrics for a recommendation feature. Pretty standard product analytics prompt but the Spotify framing threw me off a bit since I was interviewing at Pinterest.

If you were the PM for a personalized music recommendation feature (like Spotify's Discover Weekly), what metrics would you track?

“The Spotify framing was a little disorienting in a Pinterest interview.”

View Post

Pinterest·Data Scientist·Technical Phone Screen

Jun 2026

Pinterest data scientist interview with a coding problem that looked straightforward until I actually had to think about the complexity constraints. The sieve angle was a bit unexpected for a DS role, but I guess they wanted to see if you could write clean, efficient code under pressure.

Write a function that returns the first n prime numbers in ascending order, using an efficient sieve approach with a tight upper bound estimate rather than trial division. Include unit tests and discuss the time and space complexity.

“My first instinct was trial division and I basically started going down that path before catching myself.”

View Post

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

Jun 2026

Pinterest ML infra interview, system design round focused entirely on GPU memory and batch size optimization for model serving. Pretty deep technically, more math-heavy than I expected for a design question.

You have a GPU with 60 GB of memory. Walk through how you'd figure out the optimal batch size for serving a model, covering parameter memory, activation memory, KV cache for transformer models, framework overhead, and the latency vs throughput trade-off. Also explain how you'd validate your choice empirically.

“This is a single question but it's really like five questions stitched together.”

View Post

Pinterest·Data Scientist·Onsite - Behavioral / Leadership

Jun 2026

Pinterest onsite for a Data Scientist role. The loop was heavy on behavioral questions across multiple panels, hiring manager included. Lots of 'tell me about a time' stuff testing ownership, analytical judgment, and how you handle pressure. Nothing too surprising but you need real stories ready.

Tell me about a time you influenced a major decision without having direct authority over the people involved.
What's the hardest question a stakeholder ever pushed back on about your analysis, and how did you handle it?
Describe a time you were put on the spot for an example you didn't actually have prepared. How did you handle the blank?
Walk me through a project that failed or didn't deliver what you promised. What went wrong and what would you do differently?
Tell me about a time you faced an extremely high-pressure or demanding situation. What did you do?
Other than Pinterest, what mobile apps do you use most and what do you like about them?

“This one tripped me up a little because my first instinct was to pick a story where I was technically the lead.”

View Post

Pinterest·Software Engineer·Technical Phone Screen

Jun 2026

Pinterest SWE interview focused almost entirely on a heap-based column assignment problem. The follow-ups kept coming and I was not fully prepared for how deep they'd go on the design side.

Given k columns and an array of post heights, assign each post to the column with the smallest current total height (ties broken by smallest index). Implement this and return either the chosen column index per post or the final column heights.
Walk through the data structures and algorithm needed to achieve O(log k) time per post insertion.
Analyze the time and space complexity of your solution.
Design a streaming API with an addPost(h) method and a peekMinColumn() method that follow the same assignment rule.
How would you extend the design to support removing an arbitrary post and updating a post's height?
With k up to 100,000 columns and up to 1,000,000 posts, what memory layout and optimization choices would you make?
How do you handle tie-breaking rules, stability guarantees, and potential 64-bit integer overflow in column heights?

“I got the greedy logic pretty fast, min-heap keyed on (height, index).”

View Post

Pinterest·Data Scientist·Technical Phone Screen

Jun 2026

Pinterest data scientist interview with a brutal single SQL problem covering four sub-tasks at once. Window functions, rolling retention, ARPU by channel, edge cases, all in one script. Left feeling like I'd either nailed it or completely missed something obvious.

Given three tables (users, sessions, orders), write a single SQL script using CTEs and window functions that computes 7-day rolling retention for each calendar day in August 2025: among users who signed up on or before day D, what fraction had at least one session in the 7-day window ending on D?
For each user, compute their first purchase date, the number of days between signup and first purchase, and prove via ROW_NUMBER or an equivalent window approach (not MIN subqueries) that the sum of orders strictly before the first purchase is zero.
Compute the top 3 acquisition channels by 7-day ARPU over the window August 26 to September 1, 2025, where ARPU is total order revenue divided by number of users from that channel who had at least one session in that window.
Handle edge cases within the same SQL script: users with no sessions, multiple sessions on the same day, multiple orders on the same day, users who signed up after the analysis window, and UTC timezone assumptions for all timestamps. Where in the query does each case get handled?

“This is the part that tripped me up most.”

View Post

Pinterest·Machine Learning Engineer·Technical Phone Screen

Jun 2026

Interviewed for an ML engineer role at Pinterest, got a technical phone screen with a pretty fundamental deep learning question. Short round, nothing crazy, but it reminded me how easy it is to fumble the basics when you're nervous.

Where in a neural network do vanishing gradients typically occur?

“I knew this cold but still stumbled explaining it clearly.”

View Post

Pinterest·Software Engineer·Technical Phone Screen

Jun 2026

Pinterest coding screen with a grid traversal problem. Pretty straightforward on the surface but they wanted you to actually talk through complexity, not just code it up.

A robot vacuum on a 2D grid can move in 8 directions including diagonals. Starting from a given cell, find all cells reachable from that position assuming no obstacles, and return the count or a marked grid. Implement using BFS or DFS and discuss time/space complexity.

“Went with BFS because I always reach for queues on grid problems, not sure that was the right call to lead with.”

View Post

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

Jun 2026

Pinterest ML Engineer interview focused entirely on designing a Trust and Safety content moderation system. It was a deep, sprawling system design session that touched basically every layer of the stack, from data pipelines to adversarial robustness. Dense question, a lot of ground to cover in one go.

Design an ML system for detecting policy-violating content (spam, hate speech, NSFW, harassment, misinformation) on a large platform. Walk through problem framing, data pipeline, modeling, serving, human review, metrics, and adversarial robustness.
How would you structure the data pipeline for training a content moderation model, including handling label noise and policy drift over time?
What features and model architecture would you use for multimodal content moderation across text, images, and video, including account and graph-level signals?
How would you design the serving layer for a content moderation system, covering pre-publish versus post-publish moderation and latency constraints?
How would you design the human review queue and prioritization logic for a content moderation system?
What metrics would you use to evaluate a content moderation system, including fairness across languages and demographics?
How would you make a content moderation model robust against adversarial evasion attempts like obfuscation, leetspeak, or image perturbations?

“This is basically seven questions wrapped into one.”

View Post