Google DeepMind Research Engineer Interview Guide

The short answer: Google DeepMind publishes four interview stages: initial interviews with a 30-minute recruiter call, two or three skills interviews with potential peers, final interviews with team leads and leadership, then a decision. It says the process takes 4 to 10 weeks, most interviews are virtual, and you may use AI to prepare but not during live interviews or interview tasks unless told otherwise. Candidate reports say research engineer skills interviews include coding rounds where the code must run and ML rounds on fundamentals, maths, probability and ML design. Older reports describe a factual maths, ML and computer science quiz, and reports disagree on whether it is still used. Last reviewed October 2026.

Google DeepMind research engineer interview: quick answers

The Google DeepMind research engineer role

What Google DeepMind research engineers work on, the skills its job postings ask for, and what a week can look like. Google DeepMind posts the role as Research Engineer, <team>, DeepMind (for example Gemini Large Scale Pretraining, Robotics, Polaris), with Senior, Staff and Senior Staff titles and some combined Research Scientist/Engineer roles.

Google DeepMind research engineer levels and pay

Your level sets the bar you are interviewed against and the pay band of the offer.

Google Careers shows title prefixes (Research Engineer, Senior, Staff, Senior Staff) but no level codes on these postings. Ranges overlap across titles, so the table lists postings.

Pay in Google DeepMind's US job postings

Google DeepMind's US research engineer roles are posted on Google Careers. Each lists a base salary range plus a bonus target and equity.

Bonus targets seen were 15% for Research Engineer and Senior Research Engineer, 20% for Staff and some Research Engineer titles such as Responsibility and the Tech Lead Manager role, and 25% for Senior Staff. The Polaris posting also lists total compensation of $550K to $1.1M. No London or other non-US ranges are published. Ranges seen October 29, 2026.

Google DeepMind research engineer interview process

Google DeepMind publishes its interview stages on its careers page and in its interviewing guide, and says exact steps differ by role. The content of the research engineer rounds comes from candidate reports.

Based on Google DeepMind's own pages and candidate reports.

DeepMind says the whole process takes 4 to 10 weeks depending on the role, possibly longer, and that candidates sign a standard NDA electronically before interviewing. Invited candidates get prep information for each stage of their role. Last reviewed October 2026.

What is different at Google DeepMind

• DeepMind defines the role. It describes research engineers as experimentalists who bridge theory and implementation, designing and scaling systems to test new ideas, which often leads to co-authoring papers. Research Scientists normally hold a PhD and form new hypotheses and algorithms, and Software Engineers build and maintain software for research.
• The quiz is a known feature. Candidate reports from 2021 and 2022 describe a factual quiz on maths and statistics, ML and computer science, mostly theory with some derivations. Reports disagree on whether it still runs, and 2026 candidate guides describe ML depth rounds instead.
• Interviews are competency-based. DeepMind recommends the STAR method and says every candidate is scored on clearly defined criteria.
• AI is for prep only. DeepMind says you may use AI to prepare, but not during live interviews or interview tasks unless told otherwise.
• US roles are posted on Google Careers with Google-style pay: a base range, a bonus target and equity. DeepMind runs its own interview process, separate from Google's.
• Polaris is open to people without degrees. The Research Engineer, Polaris posting has no education or prior work requirement and prefers competitive programming or maths olympiad experience. Its engineers hand-build long-horizon software engineering eval tasks for Gemini, each taking about a week.

Google DeepMind research engineer interview questions

Questions from our bank, then practice prompts we wrote for this guide around Google DeepMind's work.

1from typing import Dict, List
2
3def pass_at_k(results: Dict[str, List[bool]], k: int) -> float:
4    """Mean unbiased pass@k over tasks, rounded to 4 decimals."""
5    pass

1from math import comb
2from typing import Dict, List
3
4def pass_at_k(results: Dict[str, List[bool]], k: int) -> float:
5    if k < 1:
6        raise ValueError("k must be at least 1")
7    if not results:
8        return 0.0
9    total = 0.0
10    for task, samples in results.items():
11        n, c = len(samples), sum(samples)
12        if n < k:
13            raise ValueError(f"task {task} has {n} samples, fewer than k={k}")
14        # comb returns 0 when n - c < k, so the task scores 1.0
15        total += 1 - comb(n - c, k) / comb(n, k)
16    return round(total / len(results), 4)
17
18# Complexity: O(total samples) time to count passes, plus the binomial terms; O(1) extra space.

How to prepare

A plan for the Google DeepMind loop, most important first:

1. Practice medium to hard algorithm problems in Python until you can write code that runs on the first try, since candidate reports say it must run.
2. Review maths and statistics, ML theory and CS fundamentals for factual questions and short derivations, in case your loop has a quiz-style round.
3. Practice ML design and ML coding: losses and gradients, training loops, RL estimators and how you would test a research idea at scale.
4. Prepare STAR stories with numbers on impact, mapped to the job description, since DeepMind scores competencies on defined criteria.
5. Read the DeepMind blog and recent papers from your target team, and try Gemini in Google AI Studio.
6. Practice without AI tools, since DeepMind allows AI only for preparation unless you are told otherwise.

Practice path

Courses that cover the rounds above:

• LLM Architecture
• Deep Learning
• PyTorch
• ML System Design
• GPU Programming