Company interview guide
Google DeepMind Machine Learning Interview Guide (2026)
Algorithmic coding and broad maths, stats and ML fundamentals, then team, research and culture conversations.
Google DeepMind's machine learning interview loop has 6 reported stages (Recruiter and hiring manager calls, Coding interviews, ML fundamentals / technical quiz, ML design, Team and research interviews, People and culture); typical length: 4-10 weeks.
Roles covered: Research Scientist, ML Engineer and Deep Learning Engineer. The question bank holds 5 questions reported from Google DeepMind interviews, 1 of them free to solve in your browser.
Merged with Google Brain. DeepMind and Google's Brain team merged into Google DeepMind in April 2023. Its research and engineering loops are separate from Google's general engineering loop.
- Interview process length
- 4-10 weeks
- Stages
- 6
- Reports linked to bank questions
- 5
- Bank report dates
- Oct 2021–Sep 2026
- Questions in bank
- 5 (1 free)
- Last reviewed
- October 2026
How the Google DeepMind interview loop works
A recruiter call, often with the hiring manager or team lead, then a skills stage of two to three technical interviews, then final interviews with team leads, researchers and a people and culture or program manager contact. Before about 2022 the skills stage opened with a two-hour quiz on CS, maths, stats and ML; in 2022-2023 it was usually three separate hour-long maths, ML and coding rounds; since 2024 it is most often one or two coding rounds plus one ML fundamentals round, with ML design and a lead-researcher interview in the final stage. Research-scientist loops can run to eight or more conversations, sometimes with a talk or a timed take-home. Silences of several weeks between stages are common.
The Google DeepMind interview process, stage by stage
01Recruiter and hiring manager calls
30 min recruiter call; in many loops a separate or joint chat with the hiring manager or team lead
A conversation about your background, what you want to work on and why DeepMind, with time for your own questions. Some candidates were stopped right after this stage because their interests or preferred location did not match the team, so it already screens for fit.
What it tests
- A clear, specific answer to why DeepMind and why this team
- What you would like to work on if you joined
- Whether your past work matches the team's research direction
- Logistics: location, start date, other processes in flight
How to prepare
- Read the team's recent papers and blog posts and name one or two you would build on.
- If you are also interviewing at Google or another Alphabet company, tell the recruiter; the processes can be coordinated.
- Ask which skills interviews you will get and request the preparation material for each.
Described in 27 of 53 candidate reports (5 since 2024); about 5 add an early hiring manager or team lead chat. The company's careers page describes a 30-minute recruiter call.
02Coding interviews
1-2 x about 60 min of live coding in a language of your choice, often scheduled alongside the ML fundamentals round
Recent reports describe medium to hard algorithmic problems with many follow-ups on trade-offs, and several candidates were cut here. Some problems have a mathematical or ML flavour. Older loops opened with a discussion of language internals and then asked you to implement an algorithm the interviewer described, or to read and improve a piece of code.
What it tests
- Data structures and algorithms: hash maps, trees, graphs and DFS, string parsing
- Turning a described algorithm, sometimes numerical, into correct working code
- Language internals: Python versus C++, typing, memory management, how a dict works
- Code comprehension: explaining and improving code you did not write
- Managing time across follow-up questions
How to prepare
- Get a correct solution down first, then discuss complexity and the follow-ups; running out of time is noted in feedback.
- Be ready to explain how a hash table handles collisions and resizing, and how you would build one from scratch.
- Write clean, class-based code with small functions; one candidate was caught out expecting a functional style.
- Expect two interviewers in the room at times, and practise coding while explaining aloud.
Coding appears in 34 of 53 candidate reports and in 12 of the 15 since 2024 that describe technical rounds; 7 report two coding rounds. The company's careers page describes two or three skills interviews.
03ML fundamentals / technical quiz
Since 2024 usually one 60-min ML round; 2022-2023 often separate 60-min maths and ML rounds; before 2022 a 2-hour quiz with two interviewers
Rapid, definition-heavy questions across linear algebra, calculus, probability, statistics, optimisation, ML and CS. Recent ML rounds go deeper into losses, optimizers, regularization, transformers and post-training, and may ask for the maths behind an LLM concept or how to train a very large model.
What it tests
- Bayes theorem, conjugate priors, the central limit theorem and hypothesis testing
- Convexity, gradient descent versus Newton and other second-order methods, Taylor series
- Matrix inversion, eigenvalues and positive definiteness; simple integrals
- Losses, optimizers, regularization, precision versus recall, backprop
- Transformers, fine-tuning and RL post-training; training models too large for one accelerator
How to prepare
- Practise giving a precise one-sentence definition, then a short example, for every core concept; interviewers move on quickly.
- Revise undergraduate maths and statistics broadly rather than one topic deeply; breadth is what reports complain about.
- Stay calm if an interviewer challenges a correct answer; restate your reasoning briefly and move on.
Free reported questions for this stage
- Memory of a Sharded Array on a Device MeshDeep LearningEasySolve
A maths, stats or ML knowledge round appears in 35 of 53 candidate reports (10 since 2024); format moved from one long quiz to separate hour-long rounds. The careers page describes skills interviews only in general terms.
04ML design
One round in many research engineer and ML engineer loops, usually in the final stage and run by the hiring team
When this round appears, it is often a problem from the team's own domain rather than a standard recommender template, and it leans more on ML than on infrastructure. It is not universal: one recent research engineer was told about system design in the prep guide but never got it, so ask your recruiter.
What it tests
- Framing an unfamiliar, domain-specific problem before choosing models
- Data, model, training and evaluation choices for an end-to-end system
- Measuring model quality with sound statistics
- Scaling training and inference beyond a single device
How to prepare
- Ask clarifying questions about the team's setting before reaching for a standard pipeline; one candidate failed by forcing a recommender template.
- Have an evaluation plan with confidence intervals ready for whatever you design.
Five research engineer or ML engineer reports from 2023-2026 describe an ML design round, one of them team-specific; one more mentions general system design and one says none was asked.
05Team and research interviews
Several 45-60 min conversations with team leads, research scientists and research engineers; research tracks may add a 1-hour talk, and one recent loop had a 3-hour timed take-home
The team tests whether you would contribute to its agenda: your past projects in technical detail, what you think of its papers and what you would work on. Some of these conversations are pointed or confrontational, and fit with research priorities is a common reason for rejection. Research-scientist loops can include a specialist interview on the technique the role really needs.
What it tests
- Technical depth on your own projects and papers, including implementation details
- Knowledge of the team's published work and the state of your field
- A concrete idea of what you would pursue at DeepMind
- Defending decisions under pushback
How to prepare
- Prepare a talk-length explanation of one project, with the failed attempts and what you would do next.
- Ask for the interviewers' names in advance and read their recent papers.
- Link your plans to the team's direction, not only to your own interests.
Team lead or researcher interviews appear in 16 of 53 candidate reports (5 since 2024); two mention a talk and one a timed take-home. The careers page describes final interviews with team leads.
06People and culture
1-2 conversations with a people and culture partner, a program manager or HR, usually last
A competency-based behavioral interview on motivation, collaboration and ways of working, sometimes with a deep dive into your CV. Motivation for DeepMind in particular, and views on AGI and its risks, come up throughout the process, not only here.
What it tests
- Why DeepMind rather than other frontier labs
- Collaboration and handling disagreement with a coworker
- How you work best and handle ambiguity
- Views on AGI, safety and responsible AI
How to prepare
- Use STAR and include numbers that show impact, as the official guide recommends.
- Expect the why-DeepMind question several times and keep the answer consistent and specific.
A culture, HR or program manager interview appears in 13 of 53 candidate reports (3 since 2024). The official interview guide recommends STAR for competency-based interviews.
Questions and themes reported most often
- Rapid-fire definitions across maths, statistics, ML and CS (25 reports, ml fundamentals / technical quiz)
- Medium to hard algorithmic coding with follow-ups (15 reports, coding interviews)
- Why DeepMind, and what would you work on here (10 reports, people and culture)
- Deep dive or talk on your own research and projects (9 reports, team and research interviews)
- Bayes theorem with a worked application (6 reports, ml fundamentals / technical quiz)
- Linear algebra basics: inverses, eigenvalues, positive definite matrices (5 reports, ml fundamentals / technical quiz)
- ML design round, sometimes set in the team's own domain (5 reports, ml design)
- How a hash table works and how to build a dict from scratch (4 reports, coding interviews)
What Google DeepMind asks about
- Maths, stats & ML fundamentals33%
- General coding30%
- Team & research fit17%
- Behavioral & motivation12%
- ML design8%
Google DeepMind interview prep checklist
- Can define and give an example of 50 core maths, stats and ML terms without hesitation
- Solved a Bayes update and a binomial power calculation by hand
- Implemented a hash table from scratch and can explain collision handling
- Coded a numerical algorithm from a verbal description, with tests
- Solved timed medium to hard graph and string problems with follow-ups
- Read the target team's three most recent papers and can discuss one critically
- Rehearsed a crisp why-DeepMind answer and two STAR stories
Tips
- Treat the fundamentals round as a breadth exam: build a one-page sheet per area (linear algebra, calculus, probability, statistics, optimisation, ML, CS) and drill quick definitions.
- Be able to derive, not just state, Bayes theorem applications, Newton's method and why convexity helps.
- Practise medium to hard algorithm problems against a timer, and leave room for follow-up questions; coding is where most recent rejections happen.
- Know how Python's core data structures are implemented and how Python and C++ differ in practice.
- Read the target team's recent papers and prepare a concrete idea of what you would work on there.
- Prepare one project for a deep technical walk-through, including what failed.
- Use the preparation material sent before each stage, but do not assume it lists only, or all of, what will be asked.
- Expect slow scheduling and long gaps; follow up with your recruiter after about a week of silence.
See the rest of the Google DeepMind loop
The later stages, the questions reported for them and how to prepare are in Pro.
Unlock with ProPractice the Google DeepMind loop
Work the stages in order: every reported question for this company, grouped by round, with solutions in the app.
Google DeepMind interview FAQ
How many interview rounds does Google DeepMind have for ML roles?
Candidates report 6 stages: Recruiter and hiring manager calls, Coding interviews, ML fundamentals / technical quiz, ML design, Team and research interviews and People and culture. A recruiter call, often with the hiring manager or team lead, then a skills stage of two to three technical interviews, then final interviews with team leads, researchers and a people and culture or program manager contact. Before about 2022 the skills stage opened with a two-hour quiz on CS, maths, stats and ML; in 2022-2023 it was usually three separate hour-long maths, ML and coding rounds; since 2024 it is most often one or two coding rounds plus one ML fundamentals round, with ML design and a lead-researcher interview in the final stage. Research-scientist loops can run to eight or more conversations, sometimes with a talk or a timed take-home. Silences of several weeks between stages are common.
How long does the Google DeepMind ML interview process take?
Reported timelines run 4-10 weeks (4-10 weeks, often longer).
What does Google DeepMind ask in ML interviews?
By reported question mix: Maths, stats & ML fundamentals 33%; General coding 30%; Team & research fit 17%; Behavioral & motivation 12%; ML design 8%. The most repeated themes: Rapid-fire definitions across maths, statistics, ML and CS; Medium to hard algorithmic coding with follow-ups; Why DeepMind, and what would you work on here.
How should I prepare for the Google DeepMind ML interview?
Treat the fundamentals round as a breadth exam: build a one-page sheet per area (linear algebra, calculus, probability, statistics, optimisation, ML, CS) and drill quick definitions. Be able to derive, not just state, Bayes theorem applications, Newton's method and why convexity helps. Practise medium to hard algorithm problems against a timer, and leave room for follow-up questions; coding is where most recent rejections happen.
Are there free Google DeepMind interview questions?
Yes. LeetML has 5 questions reported from Google DeepMind interviews, and 1 of them are free to solve in the browser; they are listed by stage above.
Related interview guides
- GoogleClassic algorithm rounds plus ML design, ML domain depth and Googleyness, judged by a hiring committee.
- OpenAIPractical coding, hands-on ML debugging and LLM system design, with a strong mission check.
- AnthropicA timed, level-by-level build exercise, practical live coding, LLM infrastructure design and a values interview that really counts.
- MetaFast classic coding, an AI-assisted coding project and product ML design for recommendations and integrity.
- xAIA fast loop run by engineers: a short screen, production-style coding, design, and a deep dive on your own work.
- Mistral AIAttention coded from scratch in PyTorch, a heavy LLM fundamentals quiz and practical work against the Mistral API.