Company interview guide

Coinbase Machine Learning Interview Guide (2026)

Cognitive and culture tests up front, hard SQL, live modeling on messy data, and a work-trial presentation for data scientists.

Coinbase's machine learning interview loop has 9 reported stages, from recruiter call to team fit and offer review; typical length: 2-12 weeks.

Roles covered: Data Scientist and ML Engineer. The question bank holds 10 questions reported from Coinbase interviews.

Try a free question from this loop's topics

Interview process length
2-12 weeks
Stages
9
Reports linked to bank questions
11
Bank report dates
May 2018–Jun 2026
Questions in bank
10 (0 free)
Last reviewed
October 2026

How the Coinbase interview loop works

Application review, a recruiter call and a roughly 30-minute cognitive and culture assessment (some candidates take it before the recruiter call), with a coding assessment added for engineering-style roles. Data Scientists then have a SQL-focused technical screen and a final round of about four interviews covering product and statistics cases, coding, fit and a work-trial presentation. ML Engineers report three rounds of 45 to 60 minutes: SQL plus light ML coding, a pair-programming modeling task in a notebook, and a hiring manager behavioral. Every offer is reviewed by senior leadership before it goes out, and several candidates report long silences after the final round.

The Coinbase interview process, stage by stage

01Recruiter call

About 30 min call

A scripted conversation about your background, why Coinbase and how you fit the team, plus an overview of the remaining steps. Coinbase says recruiters also probe alignment with its cultural tenets and your interest in crypto, and share level and pay range.

What it tests

  • Why Coinbase, and what you know about the company and crypto
  • Your most impactful project and how it maps to the role
  • Fit with Coinbase's stated culture and mission
  • Logistics: level, location, timeline

How to prepare

  • Read Coinbase's culture document and mission posts before the call; candidates are asked about them directly.
  • Prepare a crisp answer to why Coinbase that goes beyond the price of bitcoin.
  • Ask which rounds follow and whether your track includes a work trial or a coding assessment.

Described in 9 Data Scientist reports and 3 analyst reports (shared stage); Coinbase's interview guide lists it as step two, after application review. Several candidates took the assessment first.

02Online assessments: cognitive, culture and technical

About 30 min cognitive and culture test from a third-party vendor; ML and engineering-style roles add a timed coding assessment

Nearly everyone takes a timed test of logical, verbal and numerical reasoning combined with a culture questionnaire that asks you to rank values and react to workplace scenarios. ML Engineer candidates also report a technical assessment: one with ML multiple-choice questions plus coding tasks such as completing gradient descent and decision tree helpers, another with a standard algorithmic coding set.

What it tests

  • Speed on logical, spatial, verbal and math reasoning items
  • Consistency with Coinbase's culture on forced-choice value questions
  • ML fundamentals in multiple-choice form (ML roles)
  • Implementing ML building blocks from scratch: entropy, splits, gradient steps
  • One or more algorithmic coding problems

How to prepare

  • Take the cognitive test when you are most alert and in a quiet room; it is timed and benchmarked against internal scores.
  • Answer the culture items in line with Coinbase's published tenets; candidates describe forced rankings with no chance to explain.
  • For ML roles, drill from-scratch numpy versions of gradient descent, logistic regression and decision tree splits.

7 DS and ML reports since 2024 plus one analyst report; two ML Engineer reports add a technical test. Coinbase's guide describes the cognitive and culture test and a CodeSignal test for engineering roles.

03Technical screen (Data Scientist): SQL

One live screen with a data scientist in a shared coding environment, usually two SQL questions with an optional bonus part

The most consistent Data Scientist hurdle. Candidates call the SQL hard, with window functions over timestamps, sessions, journeys and retention; some screens also include case or statistics questions. A 2018 candidate instead got a group-expense settlement coding task.

What it tests

  • Window functions: LAG, LEAD, ROW_NUMBER and running totals over event time
  • Stitching rows into sequences, such as flight legs into journeys
  • Retention and activity metrics with a clear definition of the start event
  • Clean, correct queries under time pressure, explaining your reasoning aloud

How to prepare

  • Practise hard window-function problems until LAG and LEAD over partitions feel routine.
  • State your metric definitions (first trade or first login, calendar or rolling window) before writing SQL.
  • Budget time for the bonus part; several candidates who missed it were rejected.
  • Get used to an unfamiliar browser editor; nerves and syntax slips were common complaints.

12 Data Scientist reports describe a technical screen (6 since 2024); 6 specify live SQL with window functions, the rest give no detail.

04Domain and coding round (ML Engineer)

About 60 min live: SQL plus simple ML coding; one team used a general coding task instead

One recent ML Engineer loop opened with a domain round combining a medium SQL question that needs window functions and a subquery with a short ML coding task. A 2024 loop instead asked the candidate to build a small query API and extend it under new constraints.

What it tests

  • Ranking within groups in SQL, including tie-breaking
  • Simple ML coding without heavy libraries
  • Building and extending a small API as constraints are added

How to prepare

  • Prepare top-N per group queries with explicit tie rules; they appeared in this round.
  • Do not assume an ML title means no SQL or general coding; ask the recruiter what each round covers.

Two ML Engineer reports (2024, 2025) describe this round; contents differ (SQL plus ML coding versus an API task). Coinbase's engineering interview post also describes 60 min pair-programming coding rounds.

05Live modeling in a notebook

30-60 min pair programming in a Jupyter notebook on a given tabular dataset

The most specific ML round: you are handed a messy dataset and asked to build a working classifier with metrics while the interviewer asks questions. ML Engineer candidates and one Senior Data Scientist describe it. A recent candidate advises picking a few features and getting the simplest model to run end to end before improving anything.

What it tests

  • Cleaning missing values, inconsistent strings and outliers quickly
  • Spotting identifier columns and leakage before training
  • Training a baseline classifier and choosing sensible metrics
  • Explaining modeling choices while coding

How to prepare

  • Get a leak-free baseline (for example, logistic regression or a tree model) running first, then iterate.
  • Practise the full loop of impute, encode, split, train and evaluate from memory, with and without scikit-learn.
  • Expect a risk or fraud flavour; recent ML teams describe tree models for risk.

Three ML Engineer reports (2024-2025) and one 2022 Senior Data Scientist report describe building a model on a given dataset in a notebook.

06Product and statistics case (Data Scientist)

Conversational case interview in the final round; some screens include a shorter case

Product sense and experimentation questions set in Coinbase's business: deciding whether to launch a campaign, designing a test to lift first trades, or explaining why retail volume fell while bitcoin stayed volatile. Interviewers look for structure and statistical grounding more than a single right answer.

What it tests

  • Experiment design: metric, randomization, bias and what to do with a null result
  • Explaining A/B testing and significance in plain words
  • Launch decisions for a campaign or feature
  • Structuring a vague business question into market, user, product and risk factors

How to prepare

  • Use a fixed structure: goal, metric, hypotheses, data, test, decision.
  • Know how crypto trading products make money and what moves trading volume.
  • Prepare a plan for an inconclusive experiment, not just a significant one.

Five Data Scientist reports (four since 2024) describe product, launch or experiment questions; the cases differ by team.

07Work trial presentation

Take-home business scenario, then a 30 min call: 10-15 min presenting, the rest questions

Coinbase includes a work trial in most non-engineering panels: you prepare an analysis or solution for a business scenario and present it to interviewers you have already met. Data Scientist candidates report it; one senior candidate said it took a few days to put together. ML Engineer reports do not mention it, which matches Coinbase's note that engineering roles do live coding instead.

What it tests

  • Turning a broad dataset or scenario into clear recommendations
  • Communicating findings concisely to a mixed panel
  • Handling follow-up questions beyond what the data covers

How to prepare

  • Keep the talk to half the slot or less, as Coinbase advises, to leave time for discussion.
  • Lead with the recommendation and its business impact, then the supporting analysis.
  • Prepare for what-if questions that your data cannot answer directly.

Four Data Scientist reports (2022-2025) describe a presentation; Coinbase's guide describes the work trial for most non-engineering roles. An analyst report mentions a 6-8 hour take-home (not counted).

08Hiring manager and culture fit

45-60 min one-to-one interviews; some loops have more than one fit round

Behavioral and resume deep-dive interviews with the hiring manager and team members, plus in some loops a separate communication and organisation round. Candidates note the same project and behavioral questions can recur across interviewers, and one was asked about comfort with 50-60 hour weeks.

What it tests

  • Deep dive on past projects and decisions driven by data
  • Handling poor data quality and ambiguous requests
  • Communication, organisation and prioritisation
  • Comfort with Coinbase's intense, mission-first culture

How to prepare

  • Prepare four or five project stories and vary the angle when the same question returns.
  • Be ready to talk honestly about workload expectations; senior candidates were asked directly.
  • Tie answers to Coinbase's tenets such as clear communication and ownership.

Eight DS and ML reports describe hiring manager or fit rounds (6 since 2024), plus one analyst report (shared stage); Coinbase's engineering interview post also describes one or two engineering manager interviews.

09Team fit and offer review

Panel feedback, possible team conversations, then leadership sign-off

Coinbase says the panel shares feedback, risks must be addressed before an offer, and the CEO or COO reviews every offer. One accepted Senior Data Scientist met specific teams late in the process; several rejected candidates describe weeks of silence at this point.

What it tests

  • Consistency of signal across the panel
  • Fit with the specific team that has headcount

How to prepare

  • Ask your recruiter for a decision timeline and follow up politely if it passes.
  • If you are interested in several teams, say so early; teams run their own pipelines.

Coinbase's interview guide describes the panel feedback and executive offer review but not team matching; only one candidate report mentions late team-fit calls.

Questions and themes reported most often

  • Live SQL, often hard, with window functions and time windows (7 reports, technical screen (data scientist): sql)
  • Timed cognitive test plus forced-choice culture questions (6 reports, online assessments: cognitive, culture and technical)
  • Build a leak-free classifier on a messy tabular dataset in a notebook (4 reports, live modeling in a notebook)
  • Present a work-trial analysis to the panel (4 reports, work trial presentation)
  • Why Coinbase, and what you know about the company (4 reports, recruiter call)
  • Design or explain an A/B test for a product change (3 reports, product and statistics case (data scientist))
  • Decide whether to launch a campaign or explain a drop in trading volume (2 reports, product and statistics case (data scientist))
  • Complete gradient descent and decision tree helpers from scratch (1 reports, online assessments: cognitive, culture and technical)

What Coinbase asks about

  • SQL & data manipulation25%
  • ML coding & modeling on data22%
  • Product sense & statistics18%
  • Behavioral & culture15%
  • Work trial presentation10%
  • General coding10%

Coinbase interview prep checklist

  • Took a timed numerical and logical reasoning practice test under real conditions
  • Solved several hard window-function SQL problems, including sequence stitching and top-N per group with ties
  • Built a leak-free baseline classifier on a messy dataset in a notebook in under 45 minutes
  • Wrote gradient descent and decision tree split helpers from scratch in numpy
  • Practised one product case and one experiment design set in a crypto trading product
  • Prepared a 10-minute work-trial style presentation and rehearsed the Q and A
  • Have a specific, non-generic answer to why Coinbase

Tips

  • Treat the cognitive and culture assessment as a real gate: candidates were dropped there before speaking to anyone technical.
  • For Data Scientist roles, make hard window-function SQL your first priority; it is the most reported technical topic.
  • For ML Engineer roles, rehearse a 45-minute notebook sprint on messy data: clean, drop identifiers, baseline, metrics.
  • Keep from-scratch ML code (gradient descent, logistic regression, decision tree splits) ready for the technical assessment.
  • Learn Coinbase's business: trading fees, retail volume swings, risk and fraud; cases and modeling tasks draw on it.
  • Read Coinbase's culture document and interview guide; recruiters and the culture test lean on them.
  • Plan a short, recommendation-first work-trial deck and leave half the slot for questions.
  • Build slack into your timeline; reports range from under two weeks to three months, with gaps after the final round.

Practice the Coinbase loop

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

Coinbase interview FAQ

How many interview rounds does Coinbase have for ML roles?

Candidates report 9 stages: Recruiter call, Online assessments: cognitive, culture and technical, Technical screen (Data Scientist): SQL, Domain and coding round (ML Engineer), Live modeling in a notebook, Product and statistics case (Data Scientist), Work trial presentation, Hiring manager and culture fit and Team fit and offer review. Application review, a recruiter call and a roughly 30-minute cognitive and culture assessment (some candidates take it before the recruiter call), with a coding assessment added for engineering-style roles. Data Scientists then have a SQL-focused technical screen and a final round of about four interviews covering product and statistics cases, coding, fit and a work-trial presentation. ML Engineers report three rounds of 45 to 60 minutes: SQL plus light ML coding, a pair-programming modeling task in a notebook, and a hiring manager behavioral. Every offer is reviewed by senior leadership before it goes out, and several candidates report long silences after the final round.

How long does the Coinbase ML interview process take?

Reported timelines run 2-12 weeks (2 weeks to 3 months (about 60 days on average, per Coinbase)).

What does Coinbase ask in ML interviews?

By reported question mix: SQL & data manipulation 25%; ML coding & modeling on data 22%; Product sense & statistics 18%; Behavioral & culture 15%; Work trial presentation 10%; General coding 10%. The most repeated themes: Live SQL, often hard, with window functions and time windows; Timed cognitive test plus forced-choice culture questions; Build a leak-free classifier on a messy tabular dataset in a notebook.

How should I prepare for the Coinbase ML interview?

Treat the cognitive and culture assessment as a real gate: candidates were dropped there before speaking to anyone technical. For Data Scientist roles, make hard window-function SQL your first priority; it is the most reported technical topic. For ML Engineer roles, rehearse a 45-minute notebook sprint on messy data: clean, drop identifiers, baseline, metrics.

Are there free Coinbase interview questions?

LeetML has 10 questions reported from Coinbase 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

Official sources