Company interview guide

Brex Machine Learning Interview Guide (2026)

Practical data science loop: statistics, an ML case on a risk model, a product framing round, light coding and values.

Brex's machine learning interview loop has 7 reported stages, from recruiter call to values, behavioral and hiring manager; typical length: Few timing reports; one candidate had the final round about two weeks after the take-home.

Roles covered: Data Scientist and ML Engineer. The question bank holds 2 questions reported from Brex interviews.

Try a free question from this loop's topics

Interview process length
Few timing reports; one candidate had the final round about two weeks after the take-home
Stages
7
Reports linked to bank questions
2
Bank report dates
Aug 2023–Jan 2025
Questions in bank
2 (0 free)
Last reviewed
October 2026

How the Brex interview loop works

A recruiter call, then a technical screen. In 2024-2025 reports the screen is one or two interviews on basic statistics and a deep dive on a past project; one 2024 candidate had a hiring-manager call at this step instead. Older loops (2019-2021) sent a time-boxed take-home data challenge of two to four hours in its place. The final stage is a virtual onsite of about four or five one-hour interviews: an ML case, coding, a product or problem-framing round with data in pandas or SQL, and values or behavioral. One 2024 candidate had the onsite split over two days, with the second half only after passing the first. This outline rests on 19 data scientist reports, most of them from before 2022.

The Brex interview process, stage by stage

01Recruiter call

About 30 min with a recruiter or the people team

An introduction to the business and the team, your background and logistics such as level. Candidates often describe recruiters as clear and helpful, and some passed on detailed feedback after later rounds. One candidate was screened out on fit right after this call, and another was told the role would be at a lower level than expected.

What it tests

  • Why Brex and why this team
  • A short account of your biggest data science challenge
  • Level, location and timeline

How to prepare

  • Ask what every later round covers; candidates say Brex shares round descriptions and sometimes sample questions.
  • Clarify the level early so a down-level does not surprise you at the offer stage.
  • Ask for feedback if you do not move on; several candidates received it.

A recruiter or people-team call opens the loop in 9 of 19 data scientist reports, including five of the six since 2024; one was rejected on fit right after it.

02Technical screen: statistics and project deep dive

1-2 interviews, about 60 min each in recent reports; one candidate had a statistics coding round

Reports since 2024 describe a screen on introductory statistics, often paired with a deep dive into a project you led. Candidates call the statistics basic, so the risk is being slow or imprecise, not facing hard theory. One 2024 candidate had a hiring-manager call on projects and analytical questions at this step, and a 2021 screen covered SQL and a business case instead.

What it tests

  • Statistics fundamentals: distributions, expected values, hypothesis tests, confidence intervals
  • Writing small statistics computations in code
  • Walking through a past project: problem, data, method, impact
  • SQL and business reasoning in some screens

How to prepare

  • Review a first statistics course end to end; candidates say the bar is accuracy on the basics.
  • Pick one project and prepare the follow-ups: why that metric, what you tried first, what you would change.
  • Expect a structured, exam-like feel and answer each part directly before elaborating.

Statistics screen in 4 of 6 reports since 2024, two of them with a project deep dive; a 2021 screen used SQL and a business case. Older loops tested statistics onsite.

03Take-home data challenge (mostly older loops)

Time-boxed, about 2-4 hours, sent soon after applying; reported in 2019-2021

Older loops opened with a take-home: join several CSV tables, compute metrics or answer a business question, and sometimes fit a simple model such as a time series. Some candidates found the last question open-ended and the instructions thin, and a few were rejected without speaking to the team. None of the reports since 2024 mention a take-home, so ask your recruiter whether it still applies.

What it tests

  • Joining and cleaning several tables with pandas or SQL
  • Computing business metrics correctly under time pressure
  • Fitting and explaining a simple model
  • Writing a short, clear summary of your choices

How to prepare

  • State your assumptions in writing for any open-ended question, since nobody is there to ask.
  • Check join keys and row counts after every merge; silent duplication is the usual bug.
  • Stay inside the time box and spend the last part on the write-up.
  • Expect follow-up questions on your submission in the onsite.

Take-home or data challenge in 9 data scientist reports, all from 2019-2021, typically 2-4 hours; none of the 6 reports since 2024 mention one.

04Onsite: ML case study

About 60 min discussion, part of a 4-5 interview virtual onsite

The most specific recent prompt is inheriting a model built by someone else: what you would check first, how you would validate it and how you would own it. Older loops asked for the end-to-end process of building a model on imbalanced data, covering data selection, sampling, algorithm choice, training and deployment. Risk and fraud framing fits the applied ML team's work.

What it tests

  • Auditing an inherited model: data, labels, evaluation and monitoring
  • Building a classifier on imbalanced data from sampling to deployment
  • Choosing metrics and thresholds that match business costs
  • Interpreting a model and its errors

How to prepare

  • Have a checklist for taking over a model: label definition and delay, train-serve skew, metric choice, drift, ownership.
  • Practise talking through delayed labels such as chargebacks and how they bias offline evaluation.
  • Tie every modeling choice to the cost of false positives and false negatives.

ML interview in 8 data scientist reports, 3 since 2024; two recent ones asked what you would check when taking over another person's model.

05Onsite: coding

About 60 min; recent reports describe live data-structure coding, one older report an API design sketch instead

Accounts differ. One 2025 candidate found it easy and pointed to stacks, queues and basic data structures; one 2024 candidate got medium-to-hard algorithm problems; another said it was not puzzle-style. In 2021 one candidate sketched the APIs for a program that runs a game instead of coding live.

What it tests

  • Basic data structures such as stacks and queues
  • Standard algorithm problems at easy to medium level, occasionally harder
  • Designing clean interfaces for a small program
  • Explaining complexity and edge cases

How to prepare

  • Do a set of easy and medium array, stack, queue and hash map problems in Python.
  • Ask early whether the round is live coding or interface design.
  • Talk through edge cases before you write code.

Coding or engineering round in 7 data scientist reports, 4 since 2024; difficulty reports range from easy data structures to medium-hard problems.

06Onsite: problem framing and product case

About 60 min; a dataset or business scenario, worked in pandas or SQL and in discussion

You get a dataset or a product question and frame it: define the metric, manipulate the data and reason about causes. A 2023 candidate was asked how to define churn for a card customer and why customers churn; a 2025 candidate described easy pandas manipulation on given data. Older loops used a fraud dataset to discuss selection rules.

What it tests

  • Defining metrics such as churn for business customers
  • Data manipulation in pandas or SQL
  • Separating correlation from cause in product data
  • Proposing concrete product actions from the analysis

How to prepare

  • Practise defining churn and retention for a business with lumpy, usage-based revenue.
  • When a metric differs between churned and retained users, ask whether it is a cause or a symptom.
  • Keep pandas joins, groupbys and date handling fast and clean.

Product analysis, problem framing or data manipulation round in 7 data scientist reports, 2 since 2024; churn and fraud scenarios recur.

07Values, behavioral and hiring manager

About 60 min values or behavioral interview; some candidates also meet the hiring manager

A values-focused conversation about how you work, plus project experience from your resume. Several candidates mention the hiring manager directly, either as an early call or late in the process.

What it tests

  • Your most challenging project and what you learned
  • How you work with stakeholders
  • Fit with Brex's values and pace
  • Depth behind your resume claims

How to prepare

  • Prepare two or three stories with clear stakeholders, decisions and measurable results.
  • Read Brex's public manifesto and connect it to how you like to work.

Values, behavioral or hiring-manager conversations in 10 data scientist reports, 3 since 2024; content is rarely described in detail.

Questions and themes reported most often

  • Statistics basics: expected values, inference and model interpretation (6 reports, technical screen: statistics and project deep dive)
  • Project deep dive: walk through work you did before (5 reports, technical screen: statistics and project deep dive)
  • Take-home: join several CSV tables and compute metrics or fraud rules (3 reports, take-home data challenge (mostly older loops))
  • What you would check when taking over someone else's model (2 reports, onsite: ml case study)
  • Data-structure coding: stacks, queues and standard problems (2 reports, onsite: coding)
  • Problem framing with easy pandas manipulation on a given dataset (2 reports, onsite: problem framing and product case)
  • Define churn for a card customer and explain why customers leave (1 reports, onsite: problem framing and product case)
  • Build a model on imbalanced data from sampling to deployment (1 reports, onsite: ml case study)

What Brex asks about

  • ML modeling & case25%
  • Statistics & probability20%
  • Product case, SQL & pandas20%
  • Coding15%
  • Behavioral & project deep dive20%

Brex interview prep checklist

  • Answered a set of expected-value, hypothesis-test and confidence-interval questions without notes
  • Can present one past project in depth, including what you would change
  • Have a structured plan for validating and owning a model someone else built
  • Walked through an imbalanced fraud classifier from sampling to threshold choice
  • Defined churn for a usage-based card product and named causes versus symptoms
  • Solved easy and medium stack, queue and hash map problems in Python
  • Prepared two values stories with stakeholders and measurable results

Tips

  • Ask the recruiter for the description of every round; candidates say Brex shares them and sometimes gives similar example questions.
  • Review introductory statistics thoroughly; the screen is basic but graded strictly.
  • Prepare a checklist for inheriting a production model, with delayed fraud labels in mind.
  • Practise building a classifier on imbalanced transaction data and choosing a threshold from costs.
  • Keep pandas and SQL joins, groupbys and metric calculations quick and correct.
  • Practise defining churn and retention for business customers whose spend varies month to month.
  • Drill easy and medium data-structure problems in Python in case the coding round is algorithmic.
  • If you get a take-home, state assumptions in writing and keep to the time box.

Practice the Brex loop

Work the stages in order: every reported question for this company, grouped by round, with solutions in the app.

Brex interview FAQ

How many interview rounds does Brex have for ML roles?

Candidates report 7 stages: Recruiter call, Technical screen: statistics and project deep dive, Take-home data challenge (mostly older loops), Onsite: ML case study, Onsite: coding, Onsite: problem framing and product case and Values, behavioral and hiring manager. A recruiter call, then a technical screen. In 2024-2025 reports the screen is one or two interviews on basic statistics and a deep dive on a past project; one 2024 candidate had a hiring-manager call at this step instead. Older loops (2019-2021) sent a time-boxed take-home data challenge of two to four hours in its place. The final stage is a virtual onsite of about four or five one-hour interviews: an ML case, coding, a product or problem-framing round with data in pandas or SQL, and values or behavioral. One 2024 candidate had the onsite split over two days, with the second half only after passing the first. This outline rests on 19 data scientist reports, most of them from before 2022.

How long does the Brex ML interview process take?

Reported timelines run Few timing reports; one candidate had the final round about two weeks after the take-home.

What does Brex ask in ML interviews?

By reported question mix: ML modeling & case 25%; Statistics & probability 20%; Product case, SQL & pandas 20%; Coding 15%; Behavioral & project deep dive 20%. The most repeated themes: Statistics basics: expected values, inference and model interpretation; Project deep dive: walk through work you did before; Take-home: join several CSV tables and compute metrics or fraud rules.

How should I prepare for the Brex ML interview?

Ask the recruiter for the description of every round; candidates say Brex shares them and sometimes gives similar example questions. Review introductory statistics thoroughly; the screen is basic but graded strictly. Prepare a checklist for inheriting a production model, with delayed fraud labels in mind.

Are there free Brex interview questions?

LeetML has 2 questions reported from Brex interviews. The stage sections above link the free practice problems for the topics each round covers; the reported-question set itself is in the app.

Related interview guides