Company interview guide

Dropbox Machine Learning Interview Guide (2026)

SQL, pandas and A/B testing for data scientists; a multi-level coding assessment and an algorithms-plus-ML final for ML engineers.

Dropbox's machine learning interview loop has 7 reported stages, from recruiter screen to behavioral, values and project discussion; typical length: 2-8 weeks.

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

Try a free question from this loop's topics

Interview process length
2-8 weeks
Stages
7
Reports linked to bank questions
11
Bank report dates
Jan 2020–Sep 2023
Questions in bank
10 (0 free)
Last reviewed
October 2026

How the Dropbox interview loop works

A recruiter or HR screen, then a skills check: for data science a 45-minute live SQL and Python / pandas screen, for ML engineering a timed online coding assessment (sometimes two) or a one-hour coding interview. Data-science candidates then usually meet the hiring manager for about 30 minutes on A/B testing and background before a virtual onsite of five or six rounds (product, A/B testing, metrics analysis, SQL, cross-functional and culture rounds, sometimes a presentation). ML-engineering candidates report a long final day, once split over two days, mixing algorithm and ML questions, plus a behavioral interview. The company says every interview is virtual, the team decides together after a final review, and AI tools are not allowed during interviews or assessments unless you are told otherwise. Candidates describe totals from about two weeks to two months; several were ghosted or saw roles filled mid-process, so keep other options moving.

The Dropbox interview process, stage by stage

01Recruiter screen

About 20-30 min, non-technical, by phone or video

A conversation about your background, goals and why Dropbox. For ML engineers, say clearly which kind of role you want: one candidate who did well on the coding screen was later told he did not fit the full-stack opening he had been routed to. In some recent software-engineering processes this call came after the online assessment rather than before it.

What it tests

  • Why Dropbox and what you want from the role
  • How your experience maps to the team
  • Logistics: timeline, level, location

How to prepare

  • State the exact role and team type you want (data science, ML engineering, which product area) and ask the recruiter to confirm it.
  • Ask which skills check you will get (live SQL screen, online assessment or take-home) and whether any AI tool use is allowed.
  • Several candidates were reassigned between recruiters; keep your own notes on dates and next steps.

Recruiter or HR screen in 7 data-science and ML-engineering reports plus 1 recent software-engineering report. The company's hiring page lists an initial recruiter screen.

02Coding assessment (ML engineering)

Timed online assessment in CodeSignal with several parts that build on each other; some candidates had a second take-home challenge or a one-hour live coding interview instead

ML-engineering candidates report taking the same coding assessment as software engineers. Recent accounts describe implementing a class over about four levels, where each level adds features and forces you to rework earlier code (a file-system style API is a common theme). One 2022 ML candidate had two take-home challenges: short algorithm questions, then building a text editor. Another 2023 candidate was asked to parse a string.

What it tests

  • Clean, extensible code that survives new requirements at each level
  • Core data structures: hash maps, heaps, sorting, recursion
  • String parsing and careful edge-case handling
  • Speed: getting through most levels inside the time limit

How to prepare

  • Practise multi-level class-design problems (in-memory file store, key-value store, text editor) against a timer.
  • Write the first levels so they are easy to extend; later levels usually change earlier behavior.
  • Try the vendor's practice assessment to learn the editor before the real one.
  • Ask in advance whether the assessment allows its built-in AI helper; the company says AI is not allowed unless you are told so.

4 ML-engineering reports (2019-2023) plus 5 software-engineering reports from 2025-2026 on the shared online assessment. The company's pages describe a timed, multi-level coding assessment or technical screen.

03SQL and Python screen (data science)

About 45 min live with a data scientist: roughly 30 min SQL, then 15-20 min Python / pandas (R accepted by some teams); one 2021 candidate had a 1-hour online test instead

The most consistent step in the data-science loop. SQL questions use subscription-style tables: users, activities and payments, asking for monthly activity rates, revenue by user type, cumulative sums, percentages and rankings. The Python part is pandas data wrangling, such as bucketing teams by paid licenses and averaging users per tier. Candidates call the questions moderate rather than hard, so accuracy and speed matter.

What it tests

  • Joins, window functions and first-event attribution in SQL
  • Cohort activity and retention style metrics
  • Cumulative and rolling aggregates over dates
  • pandas merges, group-bys and bucketing with correct denominators

How to prepare

  • Drill SQL on a users / events / payments schema until monthly active rates and revenue splits take under ten minutes each.
  • Keep entities with zero activity in averages: start from the entity table and left join.
  • Talk through tie-breaking and date boundaries before you write the query.

Live SQL plus Python or pandas screen in 6 data-science reports from 2021-2024, usually 45 minutes; one older report had modeling questions instead. The company's hiring page lists an assessment or technical screen.

04Hiring manager screen (data science)

About 30 min with the hiring manager, before the onsite

A short call that mixes your background with an experimentation case: design an A/B test for a new feature, or define the metric for a landing-page button change and explain how you would run the test. Some managers also ask a behavioral question such as a conflict with a coworker. Candidates found these calls brief and sometimes hard to read, so lead with a crisp structure.

What it tests

  • Choosing a primary metric and its unit of analysis
  • Designing an A/B test end to end: hypothesis, randomization, sample size, duration, guardrails
  • Relating your past projects to the team's work
  • A quick behavioral story

How to prepare

  • Have a two-minute A/B test template ready and use the standard names (click-through rate, conversion) for metrics.
  • Ask one or two clarifying questions, then give a short answer and offer to go deeper.
  • Prepare one conflict story with a clear resolution.

Hiring-manager call before the onsite in 5 data-science reports (2015-2024); 3 of them describe a 30-minute A/B testing or metric case.

05Data science virtual onsite

5-6 interviews of about an hour over video with data scientists, product managers and the hiring manager

Rounds reported include product sense, an A/B testing round, technical analysis of metrics, a SQL round, quant and reasoning, a cross-functional partner round, culture and motivation, and in one loop a presentation. Questions are product-flavored: pick a favorite product and the metrics you would track, judge whether a business metric is healthy, design and read an experiment. One candidate found the rounds shallower than expected, while another rated the process difficult.

What it tests

  • Product sense and choosing metrics for a feature or product
  • Experiment design and interpretation
  • Reasoning about business metrics such as acquisition cost and lifetime value
  • SQL and quantitative reasoning under discussion
  • Presenting past analysis to a mixed audience

How to prepare

  • Prepare metric trees for Dropbox-like products: sign-ups, trial to paid conversion, team seats, retention.
  • For every metric you propose, name a guardrail and a likely confounder.
  • If a presentation is included, pick a project with a clear decision and measurable impact.
  • Review the company's values before the culture round.

Virtual onsite of five or six rounds in 5 data-science reports (2021-2025); 3 list product or A/B testing rounds. The company's hiring page says all interviews are virtual meetings with the team.

06ML engineering final round

A long final, about 5 hours in one report and split over two days in another, mixing algorithm and machine learning interviews

Only two ML-engineering candidates describe this stage. One had algorithm and ML questions at a difficulty compared to other large tech companies; the other got unusual problems: an algorithm to win a game and a concurrent scheduler, and warned that the questions did not match sample lists found online. Software-engineering reports from 2025-2026 also include system design and a project deep dive, but our ML reports do not confirm either, so ask your recruiter for the exact rounds.

What it tests

  • Algorithmic problem solving with clean, tested code
  • Concurrency basics: scheduling, threads, shared state
  • Core ML concepts: model choice, evaluation, data preparation
  • Explaining trade-offs in your own past ML systems

How to prepare

  • Practise a thread-safe or async task scheduler and be ready to discuss locking and ordering.
  • Prepare to walk through one ML project end to end: data, model, metrics, deployment.
  • Expect fresh problems rather than repeats; practise reasoning aloud on unfamiliar games or simulations.

Only 2 ML-engineering reports (2022, 2023) describe the final; both mix coding with ML or concurrency. The company's hiring page describes virtual interviews with the team.

07Behavioral, values and project discussion

One or more conversations, often an all-round or culture round in the onsite plus a separate behavioral interview for ML engineers

Interviewers ask about recent projects, goals, timeline and work style, how you handle conflict and feedback, and why Dropbox. Data-science onsites often end with an all-round interview on behavior and past projects, and one loop had separate people and culture-and-motivation rounds. The company says these interviews look at your strengths, working style and collaboration.

What it tests

  • Collaboration, conflict and receiving feedback
  • Ownership and impact on past projects
  • Motivation for Dropbox and the team
  • Working style in a distributed, virtual-first company

How to prepare

  • Prepare three or four stories with numbers: a disagreement, a failure, a project you led, a time you changed course.
  • Read the company's published values and tie at least one story to each.
  • Expect detailed follow-ups on any project you mention.

Behavioral or culture rounds in 5 data-science and ML-engineering reports plus 1 recent software-engineering report with a manager closing call. The company's hiring page describes these conversations.

Questions and themes reported most often

  • Design an A/B test for a new feature or a button change (4 reports, hiring manager screen (data science))
  • SQL on users, activities and payments: monthly active rate, revenue by first-activity type, cumulative counts (4 reports, sql and python screen (data science))
  • pandas wrangling: bucket teams by paid licenses and aggregate users (4 reports, sql and python screen (data science))
  • Multi-level class implementation in the coding assessment (file system, text editor) (3 reports, coding assessment (ml engineering))
  • Presentation or deep discussion of a past project (3 reports, data science virtual onsite)
  • Pick a product or feature and define the metrics to track (2 reports, data science virtual onsite)
  • Write a concurrent scheduler (1 reports, ml engineering final round)
  • Set up a propensity model for a sales segment (1 reports, sql and python screen (data science))

What Dropbox asks about

  • SQL & data manipulation28%
  • Product sense, metrics & A/B testing26%
  • Coding & algorithms24%
  • Behavioral & project discussion14%
  • ML concepts & modeling8%

Dropbox interview prep checklist

  • Confirmed with the recruiter which track, skills check and onsite rounds apply to you
  • Can write cohort, first-event and rolling-window SQL without hesitation
  • Can bucket and aggregate data in pandas with the right denominators
  • Have a ready A/B test design and a metric tree for a Dropbox-like product
  • Completed a timed multi-level class-design problem in an online editor
  • Practised a concurrent task scheduler and one unfamiliar game or simulation problem
  • Prepared stories for conflict, failure, ownership and why Dropbox

Tips

  • Work out which loop you are in early; the data-science and ML-engineering processes share almost nothing beyond the recruiter and behavioral rounds.
  • Data scientists: SQL on subscription data and pandas wrangling are the most consistent screen content, so make them automatic.
  • Data scientists: A/B testing appears at the manager call and again in the onsite; rehearse a full design including power, duration and guardrails.
  • ML engineers: practise multi-level coding tasks where each part changes the earlier code, in an online editor and against a timer.
  • ML engineers: be ready for concurrency questions and for unfamiliar problems rather than well-known ones.
  • Think about Dropbox's own products (storage, sharing, team plans, trials) when you prepare metrics and product cases.
  • Do not use AI tools in any interview or assessment unless the invitation says you may.
  • Follow up politely if you hear nothing for a week or two; several candidates reported long silences.

Practice the Dropbox loop

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

Dropbox interview FAQ

How many interview rounds does Dropbox have for ML roles?

Candidates report 7 stages: Recruiter screen, Coding assessment (ML engineering), SQL and Python screen (data science), Hiring manager screen (data science), Data science virtual onsite, ML engineering final round and Behavioral, values and project discussion. A recruiter or HR screen, then a skills check: for data science a 45-minute live SQL and Python / pandas screen, for ML engineering a timed online coding assessment (sometimes two) or a one-hour coding interview. Data-science candidates then usually meet the hiring manager for about 30 minutes on A/B testing and background before a virtual onsite of five or six rounds (product, A/B testing, metrics analysis, SQL, cross-functional and culture rounds, sometimes a presentation). ML-engineering candidates report a long final day, once split over two days, mixing algorithm and ML questions, plus a behavioral interview. The company says every interview is virtual, the team decides together after a final review, and AI tools are not allowed during interviews or assessments unless you are told otherwise. Candidates describe totals from about two weeks to two months; several were ghosted or saw roles filled mid-process, so keep other options moving.

How long does the Dropbox ML interview process take?

Reported timelines run 2-8 weeks.

What does Dropbox ask in ML interviews?

By reported question mix: SQL & data manipulation 28%; Product sense, metrics & A/B testing 26%; Coding & algorithms 24%; Behavioral & project discussion 14%; ML concepts & modeling 8%. The most repeated themes: Design an A/B test for a new feature or a button change; SQL on users, activities and payments: monthly active rate, revenue by first-activity type, cumulative counts; pandas wrangling: bucket teams by paid licenses and aggregate users.

How should I prepare for the Dropbox ML interview?

Work out which loop you are in early; the data-science and ML-engineering processes share almost nothing beyond the recruiter and behavioral rounds. Data scientists: SQL on subscription data and pandas wrangling are the most consistent screen content, so make them automatic. Data scientists: A/B testing appears at the manager call and again in the onsite; rehearse a full design including power, duration and guardrails.

Are there free Dropbox interview questions?

LeetML has 10 questions reported from Dropbox 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