G-Research Interview Guide: Quant Quiz, Process & Questions

The short answer

The G-Research quantitative research interview starts with an online quant quiz, followed by a technical triage discussion, deeper interviews and a leadership stage. The firm describes four one-hour technical interviews for its standard research track, or two one-hour interviews for ML-focused profiles.

Research candidates should prioritise probability, statistics, linear algebra and explaining code; engineering applicants follow a separate technical and behavioural process. Below are the actual stage distinctions, recent question topics and original worked exercises to practise.

How the G-Research interview process works

G-Research recruits researchers, machine-learning specialists and engineers to work on statistical models and the infrastructure behind them. Its published research process is unusually specific, so start with the route that matches your job title rather than treating it like a trader mental-maths interview.

Application review. Submit a CV and the requested academic and contact details. Research roles are screened for mathematical, scientific and programming strength.
Online quant quiz. Complete either the general quantitative aptitude quiz or the ML-specific assessment selected for your background.
Triage interview. Work through technical questions while G-Research assesses the depth and subject areas most relevant to your later interviews.
Further technical interviews. The standard researcher pathway typically has four one-hour interviews, including an in-depth mathematics round. ML-focused profiles typically have two one-hour interviews with ML, maths, statistics and programming.
Leadership interviews. Discuss your technical interests, research approach, collaboration and fit with the team that could employ you.

The firm's current careers guidance lists the quiz, technical and leadership phases; its earlier assessment guidance and current internship programme identify triage as the conversation between quiz and deeper interviews. This gives you a useful order for preparation without assuming every research internship or experienced hire has the same number of meetings.

Application and early screening

G-Research says research applicants can expect a status update within one week of applying. For research internships, the company describes a four-step selection process that typically takes one to two weeks, while an engineering internship has a different three-step process.

Make the work you want to discuss unmistakable on your CV: a statistical investigation, a piece of mathematical research, a reproducible ML experiment or a substantial engineering system. For the 2027 Summer Research Programme, the firm asks candidates to choose one research internship stream and considers applications across relevant opportunities when appropriate.

G-Research online quant quiz: format and topics

The first research assessment is a general quantitative aptitude quiz or an ML-specific quiz. G-Research's published preparation material names probability, statistics including ordinary least squares (OLS), linear algebra, calculus with differential equations, programming and finance as its general quantitative syllabus.

The assessment rewards the ability to choose a mathematical approach and finish it correctly, not just recall a definition. Practise translating a short problem into equations, checking assumptions and eliminating incorrect answers before spending time on a longer derivation.

How many questions and how much time?

A July 2026 London quantitative-researcher process included 10 probability, mathematics and statistics questions in 60 minutes. A separate October 2026 research-internship account described roughly 20 online maths questions, so the 10-question format should not be treated as the fixed specification for every applicant.

For preparation, try 10 mixed quantitative problems in 60 minutes, with a strict stop at the end. This is an EverythingQuant practice benchmark based on one recent research-route format, not a published firm-wide test configuration.

Topics worth prioritising

Subject	What to be able to do	A useful practice check
Probability	Compute conditional probabilities, expectations and distribution functions; distinguish independence from correlation.	Derive the distribution of a sample maximum, then check it with simulation.
Statistics and OLS	Derive an estimator, state its bias and variance, and explain when regression assumptions fail.	Derive normal equations and explain what happens with perfectly collinear features.
Linear algebra	Use matrix rank, eigenvalues, covariance matrices and positive definiteness.	Diagonalise a symmetric 2 × 2 matrix and interpret its principal directions.
Calculus	Differentiate an objective, test a stationary point and solve simple differential equations.	Minimise a parameterised quadratic, then solve a first-order linear ODE.
Programming	Explain an algorithm, reason about complexity and handle boundary cases.	Implement a rolling-window operation without rescanning every window.
Finance fundamentals	Understand returns, risk, simple payoffs and how models are evaluated.	Explain why an in-sample trading result need not survive out of sample.

Scope that matters

The topic list comes from G-Research's own quantitative-research preparation guidance, while the 60-minute and approximately 20-question formats are distinct 2026 candidate experiences. Use the test instructions sent for your specific application for permitted materials, platform and timing.

The triage interview: maths, estimation and coding

The triage conversation checks whether you can extend an initial solution when the interviewer changes the question. In its assessment guidance, G-Research describes triage as a way to select deeper interviews from maths, statistics, finance, programming and ML.

The July 2026 London quantitative-researcher process reached matrix-algebra questions, maximum-likelihood estimation for a uniform distribution and explaining a monotonic-deque algorithm. Each topic led to several follow-ups, making derivation, complexity and clear reasoning as important as the first answer.

Matrix algebra

Start from a small matrix, calculate its eigenvalues and check rank or invertibility. Then explain how changing an entry alters the result.

Statistical estimation

Write down a likelihood, optimise it and test whether the estimator is biased. Explain what information the observed data cannot identify.

Algorithm explanation

Give a simple baseline, improve its complexity and justify every invariant. Be able to trace the data structure on a short input.

Follow-up questions

Practise changing a distribution, removing an assumption or adding a memory constraint. State what breaks before proposing a repair.

For a triage rehearsal, choose one problem each from algebra, estimation and programming, then spend 15 minutes per problem answering follow-ups aloud. This 45-minute structure is a preparation drill, not a confirmed G-Research interview length.

Technical interviews: four research hours or two ML hours

G-Research states that its standard quantitative research process typically involves four technical interviews lasting one hour each, including a deep mathematics interview. For profiles better suited to machine learning, it instead describes two one-hour ML-focused interviews that still test relevant mathematics, programming and statistics.

The mathematical round can progress from a simple result to its proof, a generalisation and an explanation of its limitations. Candidates should be ready to work without a memorised template and to defend why each assumption is needed.

Mathematics and statistics

Revisit maximum likelihood, conditional expectation, matrix decompositions, optimisation and regression. For OLS, know that collinearity makes the coefficient solution non-unique and why regularisation can restore a well-posed optimisation problem.

Go one step beyond routine calculations by asking what happens when observations are dependent, data are missing or a covariance matrix is nearly singular. These extensions are a better simulation of a research discussion than memorising isolated puzzle answers.

Programming and research reasoning

Practise implementing a sound algorithm and explaining its time and memory complexity. If you propose a simulation to verify a probability result, state which quantities you would compare, how many trials you would run and what discrepancy would cause you to revisit the derivation.

For a research project on your CV, prepare the exact hypothesis, baseline, train/test split, failure modes and decision to continue or stop. A good explanation shows how you rejected a weak idea, not just how you trained the model that performed best.

How interviews differ by G-Research role

G-Research has distinct research, ML and engineering teams. The correct preparation changes considerably between a statistical researcher, an ML researcher and a C++ or infrastructure engineer.

Quantitative Researcher and Quantitative Analyst

Prioritise probability, estimation, statistics, linear algebra and applied programming. Recent London research interviews reached uniform-distribution maximum likelihood and matrix algebra, which are more informative preparation signals than generic trading-game questions.

Derivations: likelihoods, OLS, unbiasedness, variance and conditional probability.
Applied mathematics: eigenvectors, optimisation and simple differential equations.
Research habits: test a hypothesis, evaluate model assumptions and explain uncertainty.

Use the general quant quiz as your starting point, then practise a one-hour whiteboard-style mathematical discussion.

Machine Learning Researcher and ML research internships

Prepare for the ML-specific quiz and the firm's two one-hour ML-focused technical interviews. Expect the mathematical and statistical foundations to remain relevant rather than treating the track as a framework-usage test.

Methods: regularisation, generalisation, optimisation and bias–variance trade-offs.
Experimental design: leakage, time-aware validation, baselines and distribution shift.
Project depth: justify modelling choices, ablations and failure analysis with numerical evidence.

For practice, take a model that performs well on a random data split and explain precisely why it fails on a later chronological holdout.

Software Engineer, Research Engineer and C++ roles

G-Research publishes a separate engineering hiring sequence: technical interview or remote test, behavioural interview, further technical interviews and management interviews. Engineering internships instead use an online skills test followed by an assessment centre after application review.

Programming: data structures, complexity, correctness and readable implementations.
C++ or systems roles: memory ownership, concurrency, performance, debugging and safe resource management.
Team discussion: explain a design trade-off, a difficult production bug and a measurable improvement.

For an engineering practice session, implement one substantial algorithm, test its edge cases and then explain a design or performance trade-off. This is a role-aligned drill, not a claim that all G-Research engineers receive the same assessment vendor or question set.

Research internship and Summer Research Programme

The research internship route begins with an online general or ML quiz and triage, followed by discussions with potential mentors and a final senior-leadership conversation. G-Research says Summer Research Programme candidates can meet up to three potential mentors who probe mathematics, programming, statistics, ML or finance relevant to possible projects.

For a research internship interview, prepare a five-minute explanation of your strongest academic or personal project with a specific result and one failure you learned from. The October 2026 London research-internship process included discussion of the applicant's PhD work followed by second-year linear-algebra questions.

Final interviews and leadership conversations

G-Research's quantitative research process ends with leadership interviews after the technical rounds. Its published engineering process separately includes a behavioural interview and later meetings with the team, manager and functional leadership.

Be ready to explain why your research question mattered, how you tested it and what you did when an experiment failed. For engineering, pick an example where you traded off correctness, performance, maintainability or delivery time, and describe the evidence behind your decision.

Good questions to ask include how new research ideas are evaluated, how researchers and engineers collaborate, and what makes an intern or graduate successful on the team. Those questions connect naturally to the firm's research-intensive work rather than relying on generic culture talking points.

Three worked G-Research-style practice questions

These are original EverythingQuant exercises based on skills relevant to G-Research's published syllabus and recent interview topics. They are not reproduced past interview questions or a prediction of the exact assessment.

1. Statistics: maximum likelihood for a uniform distribution

Eight independent observations are drawn from a Uniform(0, θ) distribution, where θ is unknown and positive. The largest observed value is 4; find the maximum-likelihood estimate of θ and an unbiased estimator based on the maximum.

Show the solution and follow-up

For any θ at least as large as the sample maximum M, the likelihood is proportional to θ raised to the power −8. It decreases as θ increases, so the maximum occurs at the smallest allowed value: the MLE is θ̂ = M = 4.

The maximum obeys P(M ≤ m) = (m/θ)⁸ for 0 ≤ m ≤ θ, giving E[M] = (8/9)θ. Multiplying M by 9/8 therefore removes the bias, and the corrected estimate is 4.5.

A natural follow-up is why the MLE is biased even though it maximises the observed likelihood. Another is to derive the sampling distribution of M and use it to construct a confidence bound.

2. Linear algebra: eigenvectors and correlated signals

Two zero-mean random signals have covariance matrix Σ = [[2, 1], [1, 2]]. Find its eigenvalues, identify the direction of greatest variance and calculate Var(X₁ − X₂).

Show the solution and follow-up

The characteristic equation is (2 − λ)² − 1 = 0, giving eigenvalues 3 and 1. Their eigenvector directions are (1, 1) and (1, −1), so the sum direction has the larger variance.

Var(X₁ − X₂) = Var(X₁) + Var(X₂) − 2Cov(X₁, X₂) = 2 + 2 − 2 = 2. The corresponding variance for X₁ + X₂ is 6, consistent with the larger eigenvalue in the sum direction.

As a follow-up, consider what happens when the off-diagonal covariance approaches 2. The smaller eigenvalue approaches zero and the matrix becomes nearly singular, making numerical estimation less stable.

3. Programming: sliding-window maximum with a deque

Given a list of numbers and a window width k, output the maximum of every consecutive window. For [4, 2, 12, 3, 8, 7] with k = 3, the result is [12, 12, 12, 8].

Show the Python solution and complexity

Keep indices in a deque ordered by decreasing value. Remove expired indices from the front and smaller competing values from the back before appending each new index.

from collections import deque

def rolling_max(values, k):
    if k < 1 or k > len(values):
        raise ValueError("k must be within the input length")

    q = deque()
    result = []
    for i, value in enumerate(values):
        while q and q[0] <= i - k:
            q.popleft()
        while q and values[q[-1]] <= value:
            q.pop()
        q.append(i)
        if i >= k - 1:
            result.append(values[q[0]])
    return result

assert rolling_max([4, 2, 12, 3, 8, 7], 3) == [12, 12, 12, 8]
assert rolling_max([2, 2, 2], 2) == [2, 2]
assert rolling_max([-4, -1, -3], 1) == [-4, -1, -3]

Each index is inserted and removed at most once, so runtime is O(n) with O(k) extra storage. The key follow-up is to explain why discarding a smaller value is safe while the newer, larger value remains in the window.

ML follow-up to practise

A research model produces excellent results with a random train/test split but fails on a later chronological holdout. Explain how future-information leakage, regime changes and incorrectly fitted preprocessing could create the gap, then design a time-aware validation procedure.

How to prepare for G-Research interviews

Start with the next scheduled stage and the subject areas in G-Research's own guidance. A research candidate facing the quiz should spend more time solving short mathematical problems than building a trading simulator, while an engineer should prioritise implementation, debugging and systems fundamentals.

Practice session	Concrete task	Success check
1. Quant fundamentals	Work 10 mixed probability, linear-algebra and estimation questions in 60 minutes.	Derive answers without a reference sheet and mark every uncertain assumption.
2. Statistical depth	Derive the OLS solution and a maximum-likelihood estimator; explain bias and collinearity.	Handle two follow-up changes to the assumptions.
3. Programming	Implement a deque or rolling-statistics problem, then test empty, repeated and boundary cases.	State O(n) or O(n log n) complexity only when you can justify it.
4. Research or ML	Walk through a real model, its baseline, holdout design and one failed experiment.	Explain whether the improvement could survive distribution shift.
5. Mock interview	Do a 45-minute oral triage and a separate 60-minute deep technical discussion.	Communicate a solution and respond constructively to corrections.

The durations in this table are EverythingQuant practice targets, not the firm's test instructions. For ML-focused roles, replace part of the general quant set with optimisation, regularisation and validation problems; for engineering roles, replace it with coding and design exercises.

Probability interview questions Build speed on expectations, distributions and conditional probability. Programming interview questions Practise implementing algorithms and testing edge cases. Data Structures and Algorithms in Python Strengthen complexity reasoning and standard algorithm patterns.

If your statistics and probability foundations need more structure, use the Probability for Quants course. For a wider application plan across research and engineering firms, build a preparation roadmap after identifying the skills you need to improve.

G-Research interview FAQs

How long does the G-Research interview process take?

G-Research describes its research and engineering internship assessment processes as typically taking one to two weeks. Its broader quant-research interview guidance says the process is usually completed within a few weeks, with individual interview schedules arranged by role.

How many questions are in the G-Research online assessment?

A July 2026 London quantitative-researcher test contained 10 questions in 60 minutes, while a separate 2026 research-internship assessment had roughly 20 maths questions. These are useful role-specific reference points rather than one universal question count.

What score do you need to pass the G-Research quiz?

A current, public pass threshold for the general or ML quiz is not confirmed in G-Research's published recruitment guidance. Focus on solving the full practice set accurately rather than relying on historical cutoff claims from older formats.

Does G-Research use HackerRank, Codility or another test platform?

G-Research publicly confirms the general and ML quant quizzes, but its current research guide does not specify a single provider for every applicant. The invitation for your role identifies the actual platform and permitted resources.

Is G-Research primarily a quant trader interview?

No. Its main public hiring tracks are quantitative research, machine learning and engineering, so the most relevant interview preparation centres on statistical reasoning, research and programming rather than rapid market-making games.

Do you need finance experience or a PhD?

G-Research says previous finance experience is not necessary for its quantitative research and ML teams. Advanced academic work is common in research, but qualifications depend on the vacancy; demonstrate the mathematical and research depth your specific position requests.

Where can I find G-Research's own interview preparation material?

Start with its quant interview guide and recommended reading PDF. They explain the published research sequence and suggest books and coding practice resources.

Sources and editorial notes

This independent guide uses G-Research's public recruiting guidance and dated interview experiences available on 4 November 2026. The exercises are original EverythingQuant practice problems, not leaked interview questions, and any 2026 candidate-specific timings are deliberately scoped to those applications.

Ready to take your quant preparation further?