XTX Markets Interview Guide

How the XTX Markets interview process works

Researchers develop the models; exchange trading engineers build the infrastructure that processes market data and sends orders. That division shapes the interview far more than a generic “quant trading” syllabus.

The junior research and XTY sequences come from XTX-authored job descriptions. The engineering sequence is reconstructed from identifiable candidate accounts, so its precise ordering is less established. If you are applying for a dedicated engineering role, the team matters too: XTX's exchange trading developers work mainly in C++ and Python, while its shared platform engineering vacancy emphasises Go, Linux and developer tooling.

Application and first conversation

Start with the official XTX Markets careers page and choose the research or engineering team that fits your work. For a research application, show a model you designed, tested and improved, including why its result was credible. For engineering, describe a working system, the performance constraint you faced and a difficult implementation choice you made.

In a September 2023 London quantitative-research interview account, the candidate described a recruiter call followed by technical phone interviews. For the XTY Labs internship, the 2026 first stage was explicitly an introductory recruitment conversation. Prepare a two-minute explanation of one project that leads naturally into harder questions: what you predicted or built, what you measured, what failed, and what you changed.

The XTX Markets online assessment

The Yerevan vacancy specifies an online test covering statistics, probability theory, general mathematics (including linear algebra and analysis) and computer science/algorithms. The same posting states that it leads into four or five interviews with researchers.

Prepare to do the mathematics rather than merely recognise terminology. Useful tasks include deriving a conditional expectation, computing the variance of an estimator, explaining what a covariance matrix tells you, finding an eigenvalue in a small example, and tracing the running time of an algorithm. The published subject list supports those families; the examples here are our preparation recommendations, not leaked XTX test questions.

One detail worth getting right: the firm-authored junior researcher posting establishes the subjects and broad sequence. It does not identify HackerRank, CodeSignal, a minimum score or an exact test length. Treat numerical claims on unrelated prep pages as unverified unless they are tied to a comparable original candidate experience.

Quantitative researcher and machine-learning interviews

The researcher interviews combine fundamentals with explanations of how models behave on data. In a London quantitative-research account describing an April 2023 interview, the candidate encountered linear models, simple neural networks and a brainteaser, followed by a home assignment. A separate London machine-learning researcher account discussed neural networks, RNNs, transformers, time series, linear algebra and calculus.

Those accounts give you a practical revision order. First, derive ordinary least squares and explain how ridge regularisation changes the fitted coefficients. Then be ready to discuss why validation on random rows can overstate performance when observations are time-ordered. Finally, explain a neural network you know well enough to work through its gradients, training instability and behaviour under distribution shift.

As a rehearsal, take one of your own research projects and let a friend challenge its label definition, sampling method, leakage risks and out-of-sample evidence. Do this without slides. The Yerevan junior researcher vacancy emphasises economically sound ideas and lower-frequency signals; that makes research judgement, not just a polished model score, worth practising.

XTX Markets internship: the XTY Labs AI research process

For its New York XTY Labs AI Research Internship, XTX has a four-stage interview process in 2026. The archived posting targeted advanced-degree researchers, ideally PhD candidates, and described a 12–14-week internship with prospective start dates from January 2027. This is a specialised AI research programme, rather than a standard undergraduate trading internship.

1. Recruitment introduction. Discuss your research background and what makes the role a fit. Prepare a concise description of your strongest original result.
2. Dataset-based take-home. Receive a dataset and build an AI model from beginning to end. Practise forming a baseline, selecting a sound validation scheme, training the model and writing down the result and its weaknesses.
3. Research Director interview. Discuss your research history, deep-learning knowledge and how you arrive at new ideas. Revisit the decisions in your papers and experiments.
4. Final researcher interview: one hour. Meet an XTX quant researcher or XTY full-time researcher to discuss technical work and collaboration.

The take-home is the most distinctive part. For a realistic rehearsal, use a small public time-series dataset and build a documented pipeline: clean it, define the prediction target, fit a simple baseline, train a more expressive model, validate chronologically, compare performance and inspect where the model fails. As an EverythingQuant practice benchmark, aim for a reproducible three-hour first pass, then make a short written defence of each choice. The three-hour target is ours; XTX's posting did not specify an assignment time limit.

The same posting listed modern and classical machine learning, linear algebra, analysis, probability, hands-on programming and data engineering as assessment areas. It explicitly said prior finance experience was unnecessary. It also asked candidates to demonstrate serious research experience, so the standard for explaining original work is higher than for a general introductory internship.

Software engineering interviews: coding, C++ and systems

Algorithmic coding and efficient implementations are prominent in the published software interview accounts. A February 2025 developer candidate described LeetCode-hard problems and discussion of how to optimise the solution. Another 2025 candidate recalled being asked to code and analyse both time and space complexity. A separate October 2025 first-hand Blind account described several hard coding questions during the onsite, standard system design and a take-home assignment.

That is useful evidence for hard coding and follow-up analysis, without turning a single candidate's onsite into a universal two-question or 90-minute online assessment. For practice, solve a hard algorithm problem, explain the naive solution first, then improve it and walk through counterexamples. A strong session includes a dry run on repeated values, zero or empty inputs, and performance on large inputs.

For exchange trading C++ engineering, XTX's own vacancy asks for modern C++, data-structure and algorithm trade-offs, caches, memory paging, system calls and CPU execution behaviour. Make sure you can explain where a cache miss hurts a hot loop, why a memory allocation appears in a latency profile, and what can go wrong when two threads share state. Its published exchange-trading team description says C++ and Python are its primary languages.

The engineering details are not identical across teams. A 2026 Yerevan Python Developer candidate reported three technical interviews followed by a system-design/build assignment, with questions about Python memory management, the GIL, generators, asyncio and multiprocessing. XTX's London Shared Engineering posting, by contrast, emphasises Go services, Linux process and networking tools, build systems and CI/CD. Prepare against the actual role description rather than assuming every engineering job is a low-latency C++ seat.

Three XTX-style technical questions, worked through

These are original EverythingQuant practice problems chosen to exercise the probability, machine-learning and engineering skills described above. They are not represented as past XTX interview questions.

A noisy signal: conditional expectation

Suppose an unknown standardised return is Y ~ N(0, 1). You observe a noisy forecast S = Y + E, where E ~ N(0, 3) is independent of Y. What are the conditional mean and variance of Y given S = 2?

Why ridge regression shrinks the fit

You fit a one-parameter model ŷ = bx on two observations, (x, y) = (1, 1) and (2, 2). Minimise (1 − b)² + (2 − 2b)² + b². What coefficient do you obtain, and how would you evaluate this model on a time-ordered dataset?

A streaming window: update the average in constant time

Implement a Python object that returns the mean of the most recent N prices after each new arrival. A full scan on every update costs O(N); can you make the update O(1) and use at most O(N) memory?

from collections import deque

class RollingMean:
    def __init__(self, size):
        if size <= 0:
            raise ValueError("size must be positive")
        self.size = size
        self.values = deque()
        self.total = 0.0

    def add(self, value):
        if len(self.values) == self.size:
            self.total -= self.values.popleft()
        self.values.append(value)
        self.total += value
        return self.total / len(self.values)

# RollingMean(3): add 10 → 10, add 12 → 11,
# add 8 → 10, add 14 → 34/3.

For further exercises, work through EverythingQuant's probability interview problems or the Probability for Quants course. For algorithms, practise coding interview problems and data structures and algorithms in Python.

How hard are XTX Markets interviews?

Expect a high technical bar, especially in areas directly relevant to the job. The published junior quant process tests several mathematical subjects before four or five researcher discussions. Software candidates in 2025 reported hard algorithms and requests to justify optimisation choices. The AI internship adds an open-ended modelling exercise and a research conversation with a director.

The difficult part is often what happens after your first answer. An interviewer can ask you to derive a familiar formula, remove an assumption, find a counterexample, or explain why a result might disappear out of sample. Practise showing each step and correcting mistakes without losing the argument. A strong CV helps you reach a conversation; being able to defend the underlying work helps you finish it.

A focused XTX Markets preparation plan

Use the route that matches your application. The plan below is an EverythingQuant recommendation, not a description of the firm's scheduling or training requirements.

The fastest way to improve is to look at the specific errors you make. Re-derive formulas you could only recall, fix implementation edge cases you skipped and practise explaining the research choices you initially left vague. If your weakness is probability, start with the Probability for Quants foundations before adding more mock interviews.

XTX Markets interview FAQs

Sources behind this guide

Interview details were checked against XTX's careers material, role descriptions and dated first-hand candidate reports. The technical examples and timed drills above were created by EverythingQuant.

• XTX Markets careers and team descriptions.
• XTX-authored junior quantitative researcher listing, Yerevan (2026 archive).
• XTY Labs AI Research Internship hiring process (21 September 2026 archived posting).
• XTX C++ Software Engineer, Exchange Trading Development and Shared Engineering developer vacancy.
• London quantitative research / ML researcher candidate accounts (2023 interviews).
• Software engineering candidate accounts (2025); separate first-hand engineering account (October 2025).
• Python Developer and other role-specific interview accounts (2026).