Pinterest System Design Interview: 60 Min, Feed Ranking, 5 Prompts

Last updated: November 2026

Quick Answer

The Pinterest system design interview is one onsite round, which candidates report runs about 60 minutes, where an engineer asks you to design a real Pinterest surface, most often a recommendation or ranking system such as the home feed or related pins. It tests two-stage retrieval and ranking, latency budgets and how you measure success. Rehearse prompts aloud in OphyAI Interview Practice in its system design format, then use the AI system design interview copilot (Premium) for timed Mock rounds and live help on the real round.

Part	Detail
Who gets it	Experienced software and ML engineers (candidate-reported); Pinterest’s new grad page lists only “technical interviews”
Length	About 60 minutes, reported; one guide says about 30 minutes of design
Typical prompts	Home feed, related pins, search or autocomplete, notifications, ads
Core skill	Candidate generation, then ranking, then blending and measurement

What Pinterest Is Testing in This Round

Pinterest’s product is a ranking problem: its researchers wrote that Pixie-backed systems “contribute to more than 80% of all user engagement on Pinterest.” Interviewers want you to reason about that system, not draw boxes. The Pinterest interview guide covers the other rounds.

What they evaluate:

• Framing. Who the user is, which metric you move (saves, long-term retention), and which guardrail you refuse to regress.
• Funnel thinking. Billions of pins down to thousands of candidates, then a ranked page of a few dozen.
• Trade-offs. Freshness versus cost (real-time versus daily batch embeddings), latency versus model size.
• Depth on request. Caching, sharding and failure modes when the interviewer pushes.

Red flags:

• “I’d use a recommendation service” with no retrieval or ranking detail. TechScreen calls this framing a hard fail signal.
• Most of the round on storage schemas before naming a metric.
• No answer for what happens when the ranking model times out.

What to Expect

Format. One open-ended prompt, usually tied to a Pinterest product. Aced (formerly Exponent) says you get “about 30 minutes to describe how you’d build” it, then follow-ups. A 2025 IC15 candidate report and TechScreen’s 2026 guide describe a 60-minute round.

Who interviews. Pinterest says engineers run technical interviews. Candidates report the design round in the virtual onsite beside about three coding rounds and a behavioral round.

Tools. A shared drawing tool; TechScreen reports candidates losing 5-10 minutes to tool friction, so practice in one.

Variation. Some candidates report classic prompts such as a rate limiter (1point3acres, search summary only); the IC15 candidate got autocomplete.

Example Questions and How to Answer Them

Illustrative prompts, not confirmed Pinterest questions.

1. Design the Pinterest home feed

Approach: Name the objective (long-term engagement, not raw clicks). Generate candidates from several sources: topics followed, embedding neighbors of recent saves, graph random walks. Rank a few thousand with a model predicting several actions; Pinterest’s TransAct paper describes Homefeed mixing real-time action sequences with batch embeddings. Blend for diversity, freshness and ads, then name the A/B test and guardrail. Warm up with the classic news feed prompt first.

2. Design related pins with sub-100ms p99 latency

Approach: Pixie reported 60 millisecond latency on a graph of 3 billion nodes and 17 billion edges, so use graph walks plus embedding nearest-neighbor search (PinSage) as candidate sources. Precompute and cache popular pins; compute the long tail on demand with a cheap fallback.

3. Design autocomplete for Pinterest search

Approach: A sharded prefix index for candidates, ranked by popularity, personalization and freshness. Cover trending-query updates, per-keystroke latency and unsafe-suggestion filtering. Practice version: search autocomplete.

4. Design the notification decision system

Approach: Treat each send as expected value: predicted engagement minus annoyance and unsubscribe cost. Add frequency caps, and measure churn, which clicks hide.

5. Design a rate limiter (classic variant)

Approach: Clarify the limit scope, pick token bucket or sliding window, keep counters in a sharded in-memory store, and say what fails open versus closed. Practice version: rate limiter.

How to Prepare in 3 Days

• Day 1: the funnel. Read the Pixie, PinSage and PinnerFormer abstracts below, then sketch the home feed funnel from memory. Our system design interview guide covers the framework.
• Day 2: timed prompts. Run prompts 1 and 2 as 45-minute sessions aloud. After each, pressure-test your trade-offs as a Mock round in OphyAI System Design (Premium), whose Scorecard compares your board with a strong answer’s board, then redraw the design from memory.
• Day 3: mock. One full mock of prompt 3 or 4 in Interview Practice, then check the transcript for skipped metrics or failure modes.

On the live round, Coding Copilot in the OphyAI desktop app snaps the problem window or transcribes a question the interviewer asks out loud, and follow-ups remember the problem. Its overlay is excluded from screen sharing, though not from a webcam or a person in the room.

Start practicing →

Common Mistakes

• Generic prep only. A URL shortener drill will not prepare you for a ranking prompt.
• One-stage designs. Scoring every pin with a heavy model fails on latency.
• Offline metrics only. Name the online experiment and the guardrail.

Related Guides

• Pinterest interview guide (the full loop and pay)
• Pinterest CodeSignal assessment
• Pinterest data scientist interview
• Spotify interview guide
• Airbnb interview guide

Frequently Asked Questions

How long is the Pinterest system design interview?

Candidate reports and a 2026 prep guide describe a 60-minute round, longer than the roughly 45-minute coding rounds. Aced says you get about 30 minutes to lay out the design, with the rest for follow-ups. Pinterest has not published the length, so confirm it with your recruiter.

Do new grads get a system design round at Pinterest?

Pinterest’s careers blog describes the new grad path as three further one-hour technical interviews with engineers and does not mention system design by name. Candidate reports place a dedicated design round mainly in experienced-hire loops. If you are early career, ask the recruiter what each technical interview covers.

Do I need machine learning knowledge for Pinterest system design?

For product-facing and ML roles, yes, at the level of retrieval, ranking, features and A/B testing. You do not need to derive model math. Some software engineer candidates report classic distributed systems prompts, such as rate limiters, but even there, knowing why the feed needs low latency helps.

How does OphyAI work in a Pinterest system design interview?

Rehearse ranking and feed prompts aloud in the system design format of OphyAI Interview Practice, which returns a transcript and per-answer feedback. On the live round, the Live Copilot in OphyAI System Design (Premium) listens to the interviewer, reads your shared whiteboard and suggests what to say and draw next. On desktop its main window is left out of screen sharing, but not hidden from a webcam or a person in the room.

Which Pinterest engineering work should I read before the round?

Start with Pixie (graph-based candidate generation), PinSage (graph neural network embeddings), PinnerFormer (daily batch user embeddings) and TransAct (real-time action sequences in Homefeed ranking). Read each abstract and one diagram, so you can justify a design choice with them instead of name-dropping.

Sources and verification notes

Checked October 2026. Pinterest’s Medium blog, 1point3acres and LeetCode blocked fetches; papers by Pinterest engineers stand in for blog posts.

• Pinterest Careers, software engineering interns and new grads: first-party; engineers interview.
• Pixie paper (arXiv, 2017): Pinterest authors; graph size, latency, engagement share.
• PinSage paper (arXiv, 2018): Pinterest authors.
• PinnerFormer paper (arXiv, 2022): Pinterest authors.
• TransAct paper (arXiv, 2023): Pinterest authors.
• Pinterest Engineering, Introducing Pixie: first-party, blocked.
• Aced (formerly Exponent), Get a Job at Pinterest: third-party.
• TechScreen, Pinterest technical interview process 2026: third-party.
• Roundz, Pinterest IC15 interview experience: candidate-reported.
• 1point3acres, Pinterest onsite threads: candidate-reported, search summary only.