Company interview guide
Pinterest Machine Learning Interview Guide (2026)
Rapid-fire ML fundamentals, edge-case-heavy coding and embedding-centric recommendation design.
Pinterest's machine learning interview loop has 8 reported stages, from recruiter call to data scientist track; typical length: 4-8 weeks.
Roles covered: ML Engineer, Applied Scientist and Data Scientist. The question bank holds 51 questions reported from Pinterest interviews, 3 of them free to solve in your browser.
- Interview process length
- 4-8 weeks
- Stages
- 8
- Reports linked to bank questions
- 68
- Bank report dates
- Jun 2017–Jul 2026
- Questions in bank
- 51 (3 free)
- Last reviewed
- October 2026
How the Pinterest interview loop works
Two entry paths lead to the same onsite. Experienced hires usually have a recruiter call and then a 60-minute technical screen; new-grad and intern candidates usually get a 70-75 minute online assessment first, often before the recruiter call, and several of them went straight from it to the onsite. The virtual onsite runs four to six rounds over one or two days: two coding and two ML system design rounds at the core, with a behavioral or hiring-manager round. Senior loops often swap one design round for an ML practitioner, domain or probability round. Loops are usually tied to an org such as ads, recommendations or trust and safety, an extra design round is sometimes added after the onsite, and some candidates were offered a different level at the end. Feedback can be slow, so follow up with your recruiter.
The Pinterest interview process, stage by stage
01Recruiter call
15-30 min call, often after a recruiter or sourcer reaches out; early-career candidates often have it after the online assessment
A short conversation about your background, why Pinterest, level and logistics such as location and visa. The company's hiring page also lists an early hiring-manager conversation among the first steps.
What it tests
- A clear reason for wanting to work at Pinterest
- How your experience maps to the team, often ads or recommendations
- Level, location and sponsorship fit
How to prepare
- Ask which org the loop is for and whether it includes an online assessment, a practitioner round or a probability round.
- Raise visa sponsorship early; candidates have been dropped late over it.
A recruiter or HR call is described in 17 ML-role candidate reports (9 since 2024), usually 15-30 minutes on background, why Pinterest and visa; official pages list it too, as a 30-minute recruiter call.
02Online assessment
70-75 min timed test, about 10 items: ML multiple choice, a hand-computed forward pass, one coding task and two ML implementations from a template
The usual first step for intern, new-grad and some early-career MLE pipelines; experienced hires rarely mention it. The ML coding tasks give you a skeleton and helper functions and ask you to fill in a classical algorithm, sometimes without numpy. Candidates consistently say the time is tight for the amount of work, and in several new-grad loops a pass led straight to the onsite with no live screen.
What it tests
- Precision, recall and reading a confusion matrix
- Gradients, ensemble trade-offs and a small network's forward pass by hand
- Implementing bagging, decision trees, random forests, Naive Bayes, KNN or K-means from scratch
- One general coding problem, such as inserting parentheses to minimise an expression
How to prepare
- Drill bootstrap sampling, majority voting and Naive Bayes in log space until each takes under 15 minutes.
- Practise with plain Python lists as well as numpy; one variant bans numpy.
- Answer the quick quiz items fast and bank the time for the two implementation tasks.
Free reported questions for this stage
24 ML-role candidate reports (15 since 2024), mostly intern and new-grad, describe the assessment, consistently 70-75 minutes and about 10 items; the company's ML new-grad page names a CodeSignal skills assessment but gives no format.
03Technical phone screen
About 60 min on a shared editor: a few minutes on your resume or projects, a few rapid-fire ML questions, then one or two coding problems
The most reported round, mainly for experienced hires. The ML part is short and conceptual, with vanishing gradients and regularization asked again and again; one senior screen gave about 10 minutes to projects, 5 to ML and 35 to coding. The coding problem is usually medium to hard, often lightly re-skinned with ads or feed language, and interviewers push on edge cases. A few senior screens were coding only, and a few candidates had two screens before the onsite.
What it tests
- Vanishing gradients: which layers suffer, which activations help, residuals and normalization
- Overfitting, L1 versus L2 and why L1 gives sparse weights
- Learning-rate effects on an oscillating training loss
- Coding with heavy edge cases, for example rounding decimal strings without floats
- Designing a small class with good complexity, such as hierarchical permission checks
How to prepare
- Have crisp 60-second answers for the classic theory questions; there is little time for each.
- Ask about edge cases before you code; interviewers keep adding inputs like a missing integer part or a trailing point.
- For tree or hierarchy problems, consider walking up from the node to the root before building top-down structures.
38 ML-role candidate reports (24 since 2024) describe a live screen, about 27 pairing quick ML theory with coding; official pages describe a one-hour technical interview with an engineer but not its content.
04Onsite: coding
2 rounds, one or two problems each; senior loops sometimes have only one
Algorithmic coding at medium to hard level, drawn from the same pool as the phone screen. Graph traversal comes up often, along with recursion, binary search and implementing a small data structure such as a sparse matrix.
What it tests
- BFS and DFS over implicit graphs, such as grouping near-duplicate images
- Designing a sparse matrix with addition and multiplication
- Backtracking and dynamic programming on small search spaces
- Stating and achieving the right complexity, including binary-search solutions
How to prepare
- Treat graph traversal as a top priority; BFS or DFS problems appear in many screen and onsite reports.
- Write bug-free code first and optimise second; reviewers mention correctness under time pressure.
- Be ready to defend a correct answer calmly when an interviewer asks whether you are sure.
18 ML-role candidate reports (10 since 2024) describe onsite coding, nearly always two medium to hard rounds; official pages list 3-4 technical interviews for new grads and advise coding practice, without naming rounds.
05Onsite: ML system design
Usually 2 rounds of open discussion in MLE and new-grad loops; senior loops often have one
End-to-end ML systems framed around Pinterest's products. Embeddings for users and pins come up most, followed by ads click-through prediction and harmful or spam content detection. Interviewers push hardest on how you build training data and representations.
What it tests
- User and item embeddings from very long engagement histories
- Ads ranking and CTR: label construction, position and selection bias, calibration, cold start for new ads
- Content safety: spam, harmful and misleading content detection at scale
- Retrieval plus ranking architecture, latency and freshness
- Offline metrics, online experiments and their mismatch
How to prepare
- Know graph-based and two-tower embedding approaches and how Pinterest-style boards and saves can serve as signal.
- Explain training-data construction in detail: which events are positives, negatives, time splits and leakage.
- Tie the design to Pinterest surfaces such as the home feed, related pins and promoted pins.
Free reported questions for this stage
- Feed Ads CTR System: Training Data, Bias, Calibration and EmbeddingsML System DesignHardSolve
17 ML-role candidate reports (10 since 2024) describe ML design rounds, 12 of them two rounds; prompts on embeddings, ads ranking and content safety recur across years.
06Onsite: ML practitioner or domain round
One onsite round in some senior loops; content depends on the team
A deep dive on projects you owned, with questions on baselines, business impact, debugging and tuning. Some senior loops have a domain round on the team's area or a probability and statistics round in this slot, one staff loop had a product round, and one candidate's round with this name was really an ads design interview. Few detailed reports, so confirm the format with your recruiter.
What it tests
- Depth on a past project: data, baselines, metrics and measured impact
- Diagnosing a model that underperforms
- Explaining architectures you used, such as two-tower models
- Combinatorial probability in loops that include a statistics round
How to prepare
- Prepare one project end to end, including how you measured revenue or user value.
- Have two stories about model failures and how you found the cause.
Four senior or staff candidate reports describe this slot (practitioner, domain, probability or product round); a fifth used the name for a design round. An official blog post describes hour-long, open-ended ML Practitioner interviews.
07Behavioral & hiring manager
One round, often with the hiring manager; sometimes a separate hiring-manager call after the onsite
Collaboration, ownership and learning from mistakes, scored against Pinterest's published values. Expect follow-ups that test level fit; two candidates were offered a different level at the end.
What it tests
- Failures and mistakes and what you changed afterwards
- Teamwork and cross-functional work
- Why Pinterest and this team
- Level-appropriate scope and leadership
How to prepare
- Map stories to the values Put Pinners First, Aim for Extraordinary, Create Belonging, Act as One and Win or Learn.
- Prepare a genuine failure story; it is a frequent opener.
11 ML-role candidate reports (6 since 2024) mention a behavioral or hiring-manager round; the hiring page describes hiring-manager and competency-based interviews, and recruiting posts stress values-based stories.
08Data scientist track
60-min tech screen (SQL, Python or pandas, experimentation) then an onsite of about five rounds
Data scientist loops differ from the MLE loop: SQL and pandas coding, statistics, A/B testing and product or metric cases, plus a hiring-manager round. Some practice items here come from older reports.
What it tests
- SQL joins and aggregations across separate event tables
- Python and pandas data manipulation
- Setting up and reading an A/B test, including p-values
- Diagnosing a metric change and designing product metrics
- Basic probability and model fundamentals such as logistic regression
How to prepare
- Rehearse a full A/B test walk-through: metric, unit, sample size, guardrails and how you would read the result.
- Practise SQL and pandas in a single timed hour, since screens often combine both with a case question.
Free reported questions for this stage
- Maximum Non-Overlapping Interval SchedulingCoding & AlgorithmsHardSolve
26 data scientist candidate reports (12 since 2024): a 60-minute screen mixing SQL, pandas and an experiment or case in 8, a five-round onsite in 5; the official new-grad page lists an assessment, 3-4 technical interviews and an HM call.
Questions and themes reported most often
- Implement a classical ML algorithm from a template: bagging, random forest, Naive Bayes, KNN or K-means (14 reports, online assessment)
- Overfitting, regularization and L1 versus L2 (10 reports, technical phone screen)
- Vanishing gradients: near the input or the output, ReLU versus sigmoid, and fixes (9 reports, technical phone screen)
- Graph traversal coding (BFS or DFS) in screens and onsites (8 reports, onsite: coding)
- Data science screen: SQL plus pandas plus an experiment question (8 reports, data scientist track)
- OA quiz: confusion-matrix metrics and a forward pass by hand (7 reports, online assessment)
- Design user and pin embeddings, including very long user histories (4 reports, onsite: ml system design)
- Failure, mistakes and why Pinterest (4 reports, behavioral & hiring manager)
What Pinterest asks about
- General coding35%
- ML / recsys system design25%
- ML & DL fundamentals20%
- ML coding from scratch10%
- Behavioral & project deep dive10%
Pinterest interview prep checklist
- Can explain vanishing gradients and L1 versus L2 regularization in under a minute each
- Solved a string-parsing problem with full edge-case coverage without converting to floats
- Implemented bagging and Naive Bayes from scratch inside 15 minutes each
- Solved several BFS or DFS graph problems and a sparse matrix class under time
- Can design user and pin embeddings end to end, including long histories and cold start
- Can design an ads CTR model with clear training-data construction and calibration
- Prepared a project deep dive and a failure story tied to Pinterest's values
Tips
- Memorise short, correct answers to the classic screen questions: vanishing gradients, L1 versus L2, learning-rate instability and overfitting.
- Practise medium-to-hard coding with a focus on edge cases and clean classes, not just pattern recall.
- Make graph traversal second nature; BFS and DFS problems turn up in screens and onsites alike.
- Prepare a deep embedding story: graph-based versus two-tower, long sequences, cold start and how to build positives and negatives.
- Have an ads ranking blueprint ready with label construction, bias correction and calibration.
- If you get an online assessment, rehearse implementing bagging, random forests, Naive Bayes, KNN and K-means against a timer.
- Read a little about Pinterest's published recommendation research so your designs use its surfaces and signals.
- Follow up politely after each stage; slow or missing feedback is a frequent complaint.
See the rest of the Pinterest loop
The later stages, the questions reported for them and how to prepare are in Pro.
Unlock with ProPractice the Pinterest loop
Work the stages in order: every reported question for this company, grouped by round, with solutions in the app.
Pinterest interview FAQ
How many interview rounds does Pinterest have for ML roles?
Candidates report 8 stages: Recruiter call, Online assessment, Technical phone screen, Onsite: coding, Onsite: ML system design, Onsite: ML practitioner or domain round, Behavioral & hiring manager and Data scientist track. Two entry paths lead to the same onsite. Experienced hires usually have a recruiter call and then a 60-minute technical screen; new-grad and intern candidates usually get a 70-75 minute online assessment first, often before the recruiter call, and several of them went straight from it to the onsite. The virtual onsite runs four to six rounds over one or two days: two coding and two ML system design rounds at the core, with a behavioral or hiring-manager round. Senior loops often swap one design round for an ML practitioner, domain or probability round. Loops are usually tied to an org such as ads, recommendations or trust and safety, an extra design round is sometimes added after the onsite, and some candidates were offered a different level at the end. Feedback can be slow, so follow up with your recruiter.
How long does the Pinterest ML interview process take?
Reported timelines run 4-8 weeks.
What does Pinterest ask in ML interviews?
By reported question mix: General coding 35%; ML / recsys system design 25%; ML & DL fundamentals 20%; ML coding from scratch 10%; Behavioral & project deep dive 10%. The most repeated themes: Implement a classical ML algorithm from a template: bagging, random forest, Naive Bayes, KNN or K-means; Overfitting, regularization and L1 versus L2; Vanishing gradients: near the input or the output, ReLU versus sigmoid, and fixes.
How should I prepare for the Pinterest ML interview?
Memorise short, correct answers to the classic screen questions: vanishing gradients, L1 versus L2, learning-rate instability and overfitting. Practise medium-to-hard coding with a focus on edge cases and clean classes, not just pattern recall. Make graph traversal second nature; BFS and DFS problems turn up in screens and onsites alike.
Are there free Pinterest interview questions?
Yes. LeetML has 51 questions reported from Pinterest interviews, and 3 of them are free to solve in the browser; they are listed by stage above.
Related interview guides
- MetaFast classic coding, an AI-assisted coding project and product ML design for recommendations and integrity.
- SnapClassic coding plus fast, broad ML fundamentals and recommendation-heavy ML design, with values questions in every round.
- LinkedInProbability-flavoured coding, ML fundamentals and recommender design built around LinkedIn's own products.
- TikTokHour-long mixed rounds: a resume grilling, ML fundamentals and an algorithm problem, repeated two to four times.
- AirbnbMarketplace ranking design, a hard look at your past models, one practical coding round and a values check.
- RedditA live model-building screen on a small dataset, then practical coding, ML fundamentals, ranking-system design and a hiring manager round.
Official sources
- Interviewing at Pinterest (official)
- Advice from Pinterest recruiting (official blog)
- Pinterest: Interview Process: Machine Learning and Data Science Interns and New Grads (official)
- Pinterest: Inside Machine Learning at Pinterest (official)
- Pinterest: Behind the Boards: Insight from Pinterest Recruiting (official)
- Pinterest: Our Philosophy on AI in Hiring (official)