Pinterest System Design Interview Questions and Answer Outlines

Pinterest design rounds ask you to build parts of Pinterest. The home feed, visual search, image upload and storage, and the recommendation system are the questions candidates report most. The round runs about 60 minutes.

One thing separates this loop from others. Product judgement is graded next to the architecture. Saying how a design choice changes what a user sees earns real credit.

The Questions Candidates Report

A Worked Outline: Design the Home Feed

Use this order. It covers the parts the interviewer scores.

• Start with numbers. Assume 500 million users and 50 million active each day. Each feed request returns 20 pins. State the read rate before you design.
• Name the read and write balance. Feeds are read far more than written. That points to precomputing.
• Candidate generation. Pull a few thousand pins from several sources: boards the user follows, topics they browse, and popular pins in their country.
• Ranking. Score candidates with a model, then cut the list to 20. Keep the model call off the request path where you can.
• Storage. Keep pin data in a sharded store keyed by pin id. Keep the user graph separately, because it is read on every request.
• Caching. Cache the ranked list for a short window. A user who scrolls twice should not trigger two full rankings.
• Serving images. Send image URLs, not image bytes. A content delivery network serves the files from a nearby location.

Image Storage, the Second Most Common Question

Pinterest is an image product, so this question appears often. Four points cover it.

• Store the original once. Object storage holds the uploaded file. The database holds only metadata and a key.
• Generate sizes ahead of time. A phone feed and a desktop board need different widths. Make them on upload, not on read.
• Use a content delivery network. Most reads should never touch your servers.
• Deduplicate. The same image is pinned many times. Hash the file and store one copy.

Visual Search, the Question That Surprises People

Visual search finds pins that look similar to a picture. The design has three parts.

A model turns each image into a vector, which is a list of numbers describing the picture. Those vectors go into an index built for nearest neighbor lookup. A query image is turned into a vector the same way, then matched against the index.

You are not expected to design the model. You are expected to design the pipeline around it. Say how new pins get indexed and how the index is kept fresh.

What Interviewers Score

• Requirements first. Ask what scale and what latency before drawing anything.
• One clear diagram. Client, service layer, cache, storage, and the offline jobs.
• Named tradeoffs. Precompute against freshness. Cost against speed. Say which side you chose.
• Product sense. Explain what the user notices if your choice is wrong.
• Time control. Leave five minutes for failure cases and monitoring.

How to Prepare

• Learn the building blocks first. Grokking System Design Fundamentals covers caching, sharding, and queues.
• Practice complete designs. Grokking the System Design Interview includes feed, search, and media storage problems.
• Pick the right course order. See which system design course to take first.
• Know the whole loop. The Pinterest interview process round by round shows where this round sits.
• Prepare the values round too. Top Pinterest behavioral interview questions lists what the panel asks.
• Practice out loud once. Mock interviews with a working engineer expose pacing problems early.