Company interview guide
Stripe Machine Learning Interview Guide (2026)
Practical multi-part coding, a timed dataset-to-model round, an ML bug squash and payments-flavoured ML design.
Stripe's machine learning interview loop has 7 reported stages, from recruiter call to data scientist track: take-home, sql and statistics; typical length: 4-8 weeks.
Roles covered: ML Engineer and Data Scientist. The question bank holds 15 questions reported from Stripe interviews, 3 of them free to solve in your browser.
- Interview process length
- 4-8 weeks
- Stages
- 7
- Reports linked to bank questions
- 19
- Bank report dates
- Feb 2020–Apr 2026
- Questions in bank
- 15 (3 free)
- Last reviewed
- October 2026
How the Stripe interview loop works
A recruiter call, sometimes a hiring-manager screen, then the first technical step: either two live screens of about an hour each (a practical programming task and an ML coding round where you build a model from a dataset, about 2 hours in all) or a one-hour online assessment. The onsite is usually four rounds (coding, ML bug squash, ML system design, hiring manager); a few loops had a fifth round, a general algorithm round or a dataset-to-model ML coding round. Candidate reports do not give total timing; the 4-8 week figure comes from a public prep guide and is unconfirmed.
The Stripe interview process, stage by stage
01Recruiter call
About 30 min by phone or video; some loops add a mostly behavioral hiring-manager screen before any technical round
A short conversation about your background, your reasons for leaving, why Stripe and where you are with other processes. Three ML-role candidates were also passed to a hiring manager or ML manager for an early, mostly behavioral screen.
What it tests
- Why Stripe, and interest in payments and financial infrastructure
- How your experience maps to the role and level
- Where you are in other processes and your timeline
How to prepare
- Ask which screen format you will get (live coding plus ML integration, or an online assessment) and what environment to install.
- Have a two-minute summary of an ML project that shipped and what it moved, with numbers.
Recruiter or HR call first in 5 ML-role and 6 data scientist reports; 3 ML-role reports add an early hiring-manager screen. Stripe's jobs FAQ also names a recruiter screen as the first step.
02Technical screens
Usually 2 x 60 min, about 2 hours in all: practical coding in CoderPad or your own IDE plus an ML coding round in a notebook.
Several 2024-2026 candidates instead got a 1-hour HackerRank assessment (long multi-part payments or fraud prompt) or a 45-min ML training assessment
The programming round is a semi-realistic task in three or four parts that build on each other, such as filtering which services a user can see. The ML coding round hands you a tabular dataset and expects a working, evaluated model in 50-60 minutes, with little guidance from the interviewer. The online-assessment variant packs several parts behind one function with mixed test cases and unclear input types. One 2026 screen swapped the pair for a fraud design discussion and a two-target regression problem.
What it tests
- Parsing messy inputs and extending one function as each part adds a rule
- Passing every test case, including edge cases, more than raw efficiency
- Going from raw table to target, features, split, model and metric in under an hour
- Handling class skew with balanced accuracy, AUC or reweighting, and explaining the choice
- Normalizing targets or features with very different scales
How to prepare
- Get a rough end-to-end model working first, then improve it; reports stress finished over perfect.
- Know pandas and scikit-learn APIs from memory; notes may be allowed but there is little time to look things up.
- Read the long prompt twice and write down input types for each part before coding, especially on the online assessment.
- Write small tests for each part yourself; some assessments mix all tests together.
Free reported questions for this stage
- Two Regression Heads With a Negative, Heavy-Tailed TargetDeep LearningMediumSolve
16 ML-role reports describe the first technical step: 8 a multi-part live coding screen, 7 a dataset-to-model round, 6 an online assessment. Stripe's jobs FAQ also lists a technical or skills-based assessment.
03Onsite: practical coding
One round in CoderPad or your own IDE with screen share
Similar in spirit to the screen: an object-oriented, multi-part problem at roughly easy difficulty per part, where test coverage and clean structure matter. One 2024 onsite had a general algorithm round instead. No Stripe-tagged problem from this round yet, so practise with the multi-part builds linked under recurring questions.
What it tests
- Modelling a small domain with classes and clear interfaces
- Extending your own design as new requirements arrive
- Edge cases around ordering, duplicates and malformed input
- Writing and running tests as you go
How to prepare
- Keep each part's logic in its own method so the next part is a small change.
- Say which edge cases you are covering before you code them; interviewers grade on completeness.
A coding round appears in 6 ML-role onsite reports from 2022 to 2025; two call it practical or class-based, one calls it a general algorithm round.
04Onsite: ML bug squash
One onsite round in your own IDE or a prepared environment: a repository with failing tests
You get an existing codebase, sometimes a large public repository, clone it, run the tests and fix the bugs. Reports give little detail beyond that, and one candidate found the pre-interview environment setup notes did not match the real task. No Stripe-tagged problem yet; the recurring section links close practice matches.
What it tests
- Navigating unfamiliar code quickly from a failing test
- Spotting data-processing and model bugs such as wrong shapes, leakage or mishandled preprocessing
- A structured hypothesis-and-check debugging process
- Explaining the root cause, not only the patch
How to prepare
- Run the tests first and work from the first failure instead of reading the whole repository.
- Set up and test your local Python environment ahead of time, but expect the real setup to differ.
- Narrate each hypothesis and how you will confirm it, so the interviewer can follow your reasoning.
An ML debugging or bug squash round is in all 6 ML-role onsite reports from 2022 to 2025; two describe cloning a repository and fixing failing tests, one a big public repository.
05Onsite: ML system design
One onsite round, open discussion; one 2026 candidate had it in the screening stage instead
Prompts are often Stripe-specific rather than textbook recommenders; payments fraud is the one prompt a candidate named, and another said real-time deployment got real emphasis. Expect pressure on real-time scoring, delayed labels and the cost of blocking good customers.
What it tests
- Framing approve, block, challenge or review decisions against fraud loss and user friction
- Labels that arrive weeks later through chargebacks and disputes
- Low-latency online features and serving
- Thresholds, monitoring, retraining and adversarial drift
How to prepare
- Agree on the decision the model drives and its latency budget in the first five minutes.
- Leave time for deployment and monitoring; real-time serving was a reported focus.
- Tie metrics to money: fraud loss avoided versus revenue lost to false declines.
ML design is in all 6 ML-role onsite reports from 2022 to 2025 and in one 2026 screen; only that 2026 report names a prompt (fraud detection), so prompt details are thin.
06Hiring manager & behavioral
One conversation with the hiring manager or a cross-team leveler
A mostly behavioral conversation with the hiring manager about how you work and what you have delivered. ML-role reports give few details; across tracks candidates were asked about projects they delivered, conflict, ambiguity, prioritization and pushing back.
What it tests
- Ownership and impact on a project you delivered
- Working through ambiguity and prioritizing with too little time
- Collaboration, disagreement and pushing back on a request
- Why Stripe and fit with its operating principles
How to prepare
- Prepare revenue, cost or latency numbers for your main projects so impact is concrete.
- Read Stripe's published operating principles and pick stories that match them.
A hiring-manager or behavioral round is in all 6 ML-role onsite reports, and 2 more ML-role reports had an early manager screen; question detail comes mostly from data scientist reports.
07Data scientist track: take-home, SQL and statistics
Recruiter or hiring-manager screen, a take-home on payments or merchant data with a written report or slides, then a virtual onsite of 4-6 rounds of about 45 min (SQL, product or analytics case, statistics or forecasting, take-home presentation, behavioral, hiring manager)
Data scientist loops are built around a take-home that is described as 4-8 hours but which candidates repeatedly say took 10-16. Prompts centre on merchants and customers: who to target for a campaign or a new product, churn, spend frequency, customer value. Many candidates were rejected after the take-home without feedback. The probability and statistics problems below come from a public question list without round details.
What it tests
- Defining a problem from a loosely worded payments dataset and writing it up
- Finding target users or merchants without labels
- Classical statistics: MLE, OLS inference, logistic regression derivations
- Probability reasoning, for example default risk under an exponential model
- Presenting and defending analysis choices
How to prepare
- Time-box the take-home and state assumptions explicitly in the report.
- Practise deriving MLE and OLS results on paper and explaining them in plain words.
- Keep SQL sharp; several onsites include a dedicated SQL round.
Free reported questions for this stage
- Logistic Regression: Gradient, Hessian, and ConvexityStatistics & InferenceMediumSolve
- Zero Correlation Does Not Imply IndependenceProbabilityEasySolve
22 data scientist candidate reports from 2018 to 2026 describe the take-home; 6 name a SQL round and 3 a statistics or forecasting round in the onsite.
Questions and themes reported most often
- Data science take-home on payments or merchant data, then present it (22 reports, data scientist track: take-home, sql and statistics)
- Behavioral stories on delivered projects, conflict, ambiguity and prioritization (10 reports, hiring manager & behavioral)
- Build and evaluate a model from a fresh dataset in about an hour (9 reports, technical screens)
- Multi-part practical coding where each part adds a rule (8 reports, technical screens)
- Find and fix bugs in an existing repository (6 reports, onsite: ml bug squash)
- Find target merchants or users for a campaign or new product with no labels (4 reports, data scientist track: take-home, sql and statistics)
- Predict customer churn, spend frequency or future customer value (4 reports, data scientist track: take-home, sql and statistics)
- Long payments- or fraud-themed online assessment with several parts behind one function (3 reports, technical screens)
What Stripe asks about
- Practical coding30%
- ML integration (dataset to model)22%
- ML debugging15%
- ML system design13%
- Behavioral & hiring manager10%
- Stats & probability (DS track)10%
Stripe interview prep checklist
- Built and evaluated a classifier on an unseen tabular dataset in under 50 minutes
- Solved a three- or four-part practical coding task with tests for every part
- Fixed planted bugs in an unfamiliar ML codebase starting from failing tests
- Can sketch real-time fraud scoring with delayed labels, thresholds and monitoring
- Have two or three behavioral stories with concrete impact numbers
- Have a clear answer to why Stripe that connects to its operating principles
Tips
- Drill a one-hour dataset-to-model routine until it is automatic: load, define target, drop leaky columns, split, baseline, metric, one improvement.
- Use metrics that survive class skew, such as balanced accuracy, PR-AUC or ROC-AUC, and be ready to explain why.
- Practise multi-part problems in a plain editor and write your own small tests for every part.
- Rehearse debugging an unfamiliar repository: run the tests, read the traceback, form a hypothesis, check it.
- Prepare one payments-flavoured design, fraud scoring at authorization, covering delayed labels, latency and monitoring.
- Ask the recruiter whether your first technical step is live screens or an online assessment; on an assessment, read the prompt twice and note input types for each part.
- Install and test your local Python, pandas and scikit-learn setup before the interview day.
- For the data scientist take-home, time-box your work, state assumptions, and write for both a business and a technical reader.
See the rest of the Stripe loop
The later stages, the questions reported for them and how to prepare are in Pro.
Unlock with ProPractice the Stripe loop
Work the stages in order: every reported question for this company, grouped by round, with solutions in the app.
Stripe interview FAQ
How many interview rounds does Stripe have for ML roles?
Candidates report 7 stages: Recruiter call, Technical screens, Onsite: practical coding, Onsite: ML bug squash, Onsite: ML system design, Hiring manager & behavioral and Data scientist track: take-home, SQL and statistics. A recruiter call, sometimes a hiring-manager screen, then the first technical step: either two live screens of about an hour each (a practical programming task and an ML coding round where you build a model from a dataset, about 2 hours in all) or a one-hour online assessment. The onsite is usually four rounds (coding, ML bug squash, ML system design, hiring manager); a few loops had a fifth round, a general algorithm round or a dataset-to-model ML coding round. Candidate reports do not give total timing; the 4-8 week figure comes from a public prep guide and is unconfirmed.
How long does the Stripe ML interview process take?
Reported timelines run 4-8 weeks.
What does Stripe ask in ML interviews?
By reported question mix: Practical coding 30%; ML integration (dataset to model) 22%; ML debugging 15%; ML system design 13%; Behavioral & hiring manager 10%; Stats & probability (DS track) 10%. The most repeated themes: Data science take-home on payments or merchant data, then present it; Behavioral stories on delivered projects, conflict, ambiguity and prioritization; Build and evaluate a model from a fresh dataset in about an hour.
How should I prepare for the Stripe ML interview?
Drill a one-hour dataset-to-model routine until it is automatic: load, define target, drop leaky columns, split, baseline, metric, one improvement. Use metrics that survive class skew, such as balanced accuracy, PR-AUC or ROC-AUC, and be ready to explain why. Practise multi-part problems in a plain editor and write your own small tests for every part.
Are there free Stripe interview questions?
Yes. LeetML has 15 questions reported from Stripe interviews, and 3 of them are free to solve in the browser; they are listed by stage above.
Related interview guides
- PayPalSQL and Python screens, medium-difficulty algorithm coding, ML fundamentals and fraud-flavoured cases, with loops that vary by team.
- Block (Square)Live pairing on practical Python and SQL, then a long onsite of hands-on modeling, case studies and ML Q&A.
- CoinbaseCognitive and culture tests up front, hard SQL, live modeling on messy data, and a work-trial presentation for data scientists.
- AffirmPractical, job-shaped ML interviews: EDA, feature engineering, metrics for credit and fraud models, plus medium-level coding.
- Capital OneA timed four-question coding assessment, then a four-round Power Day of ML technical, a business case, behavioral and, on recent ML engineer loops, ML system design.