Articles

G-Research Software Engineering Internship 2027: Coding Assessment, Mathematics, and Interviews

G-Research Software Engineering Internship 2027: official assessment stages, coding and mathematics topics, interview preparation, and timeline guidance.

If you are targeting the G-Research Software Engineering Internship 2027, prepare for a compact process with a coding or skills screen and an assessment centre, while keeping the exact test format flexible. G-Research's current internship page says 2027 engineering applications are not open yet and asks candidates to register interest. Its current student guide says the general engineering process has three stages and typically takes one to two weeks: apply online, complete an online skills test, and attend an assessment centre. That is the reliable starting point; a HackerRank timer, question count, cutoff, or exact interview schedule is not publicly guaranteed.

Show that you can turn an ambiguous problem into tested code, explain complexity, and revise assumptions when requirements change.

What G-Research confirms for 2027

The current Engineering internships page describes 12- or 24-week software engineering and FPGA engineering internships in central London. It says engineers build the platforms, infrastructure, and tools that let researchers test ideas at scale, including technology for difficult quantitative-finance problems. The software-engineering stream contributes to systems and platforms supporting quantitative research and asks interns to improve and modernise high-performance technology.

The same page currently says, "Register your interest for our 2027 engineering internships and we'll let you know when applications open." That wording matters. You can prepare now, but do not treat a third-party listing, an old deadline, or a generic G-Research vacancy as proof that the 2027 software-engineering application is live.

G-Research's Students and graduates guide adds role context. Engineering interns spend 12 or 24 weeks embedded in Quantitative or Infrastructure Engineering, working on trading-technology platforms, legacy systems, or OpenStack work. For preparation, this points toward production-quality habits: readable interfaces, defensive tests, performance awareness, and the ability to learn an unfamiliar codebase.

The official assessment timeline

G-Research publishes a high-level process rather than a detailed 2027 playbook. The engineering section lists three stages and says the process typically takes one to two weeks to complete.

Stage	What the official page says	What to prepare	What remains unknown
1. Apply online	Submit an application for the engineering internship.	A concise project narrative, coursework, and links to code you can explain.	Opening date, eligibility filters, and response time for your application.
2. Online skills test	Complete an online skills test.	Timed data structures, algorithms, implementation, debugging, and complexity explanations.	Vendor, number of tasks, timer, language rules, proctoring, and cutoff.
3. Assessment centre	Attend an assessment centre.	Live coding, design discussion, project deep dive, collaboration, and questions for engineers.	Remote or in-person format, number of sessions, and whether there is a separate manager round.

The one-to-two-week figure is a typical duration, not a service-level promise. Scheduling and team demand can lengthen it. If your invitation conflicts with a public guide, follow the invitation and ask the recruiter one precise question about the next stage.

How to read candidate reports without overfitting

Public reports help choose drills, but they span different years and teams. The G-Research Software Engineer(Internship) page on Glassdoor was updated in November 2021. One report described three easy algorithmic questions on HackerRank and an option to interview or solve tasks at home. A 2020 report described a remote assessment centre with group design and individual interviews. Another described four HackerRank tasks, two easy and two medium, with conflicting timing details.

Those details disagree because they are historical, anonymous reports. They do not establish the 2027 provider or timer. A Dataford synthesis (updated August 2026) groups interviews into recruiter, online assessment, coding, system design, and behavioral themes. InterviewSense examples dated April and May 2026 mention arrays, iterative trees, merge intervals, rate-limiters, and project discussions, while cautioning that its data may not reflect the exact process.

The safe inference is narrower: practise the underlying skills and expect the format to vary. The unsafe inference is "G-Research always sends a 90-minute HackerRank test" or "every intern completes six rounds." Your own invitation is the authority.

Coding assessment: optimise for correctness you can explain

For a skills screen, start with a small contract before typing. Write down input constraints, output shape, duplicate behavior, ordering, and failure cases. Then choose a data structure and state the expected time and space costs. This makes a medium problem easier to review and gives an interviewer something concrete to challenge.

The likely practice surface is broad but familiar:

Arrays, strings, hash maps, sorting, binary search, and intervals.
Trees and graphs, including iterative traversal and cycle or dependency handling.
Dynamic programming when a problem has overlapping subproblems and a clear state.
Testing, parsing, error handling, and deterministic output.
Basic performance reasoning: memory layout, avoidable copies, and the cost of an extra pass.

Do not spend preparation time guessing whether the assessment uses HackerRank, CodeSignal, or another provider. Build two timed simulations in your preferred language, then repeat each solution without hints. Be ready to explain why a hash map fits the key domain, why sorting is acceptable, and what changes when data arrives as a stream.

A worked diagnostic: from intervals to a production follow-up

Suppose an assessment asks you to merge overlapping intervals. A strong answer has four visible steps:

Clarify the contract. Are intervals closed, so [1, 3] and [3, 5] merge? Can an interval be empty? Must the output preserve input order?
State the invariant. After sorting by start time, the output's last interval covers every processed interval that overlaps it. If the next start is beyond the last end, append a new interval.
Give the bound. Sorting costs O(n log n); the scan costs O(n); the output uses O(n) space, or O(1) auxiliary space if in-place mutation is allowed.
Test boundaries. Use an empty list, one interval, touching endpoints, nested intervals, duplicate intervals, and an interval with a negative start.

The follow-up can make the question more G-Research-like without requiring a finance formula: intervals now arrive out of order as a stream. You should say that the original one-pass invariant no longer holds globally. If the window is bounded, keep a min-heap and emit only intervals older than the watermark. If late data is allowed, define a correction policy and its latency. This answer demonstrates that you can move from a correct algorithm to an explicit system trade-off.

Practise that reasoning with Merge overlapping time intervals efficiently, then compare the stateful version in Implement sliding-window rate limiter function. Both reward a clear invariant more than a memorised trick.

Mathematics that supports software engineering

G-Research separates engineering internships from its quantitative research internship process. The engineering page does not publish a mathematics test. Still, the company's engineering work supports quantitative research, so mathematical fluency helps you reason about data, capacity, and correctness.

Focus on practical mathematics rather than trying to reproduce a quant-trader puzzle bank:

Discrete probability: conditional probability, expectation, and simple counting.
Statistics: mean, variance, quantiles, sampling error, and the difference between correlation and causation.
Numerical thinking: units, magnitude checks, overflow, floating-point error, and stable comparisons.
Complexity math: sums, logarithms, amortised cost, and how throughput changes when input or fan-out grows.
Experiment discipline: train/test leakage, a baseline, a confidence interval or uncertainty statement, and a falsifiable hypothesis.

For a spoken question, explain the assumptions before the calculation. If asked for an expected value, say what is random, what is independent, and what distribution you are using. If asked to compute a statistic from compressed data, discuss integer overflow, empty input, and whether an exact median is required. Compute Statistics from a Frequency Array is a useful implementation drill, while Solve probability and expectation problems rehearses concise derivations.

Assessment-centre interviews: show your engineering loop

The current official page names the assessment centre but does not publish its internal agenda. The historical 2018 Quant Platform interview article is useful context, clearly marked as historical: it described an at-home programming task, a three-interview technical half-day, a homework review, a design interview, hard algorithm questions, manager conversations, and a team meeting. That process may have changed, so use the behaviors rather than the sequence.

In a live round, make your engineering loop observable:

Restate the requirement and ask one high-value clarification.
Propose a simple solution and name its invariant.
Implement a small correct path before adding optimisations.
Run a normal case and an edge case aloud.
Quantify the cost and explain what new requirement would change your design.

For design, start with the request path and data ownership. A rate limiter, for example, needs a key (user, token, or IP), a policy (fixed, sliding, or token bucket), a clock, and a decision when shared state is unavailable. Then discuss atomicity, hot keys, memory expiry, and whether slight over-admission is preferable to blocking all traffic. Practise with Design a trie-based URL router with wildcards for API and data-structure design, and use the rate-limiter problem as a distributed-state follow-up.

A preparation decision tree for the next seven days

Use your evidence to choose the next drill instead of following a generic calendar:

No invitation yet? Build a one-page project story, refresh arrays/hash maps/trees, and register interest on the official G-Research page when the 2027 form is available.
Online skills test scheduled? Prioritise two timed coding sets, then review failed tests and complexity. Do not spend the session memorising a vendor's rumored question count.
Assessment centre scheduled? Add two spoken mock interviews, one design exercise, and a project walkthrough. Practise accepting a hint and updating your plan.
Asked about quantitative work? Rehearse probability, expectation, summary statistics, and uncertainty. Connect each calculation to a data or systems decision.
Waiting after a stage? Use the official one-to-two-week figure as a rough context, then send one concise status question after the timeframe in your invitation has passed.

FAQ

Is the G-Research Software Engineering Internship 2027 open?

The current engineering-internship page asks candidates to register interest and says G-Research will notify them when applications open. Check the official page for the live opening rather than relying on an old aggregator listing.

Does G-Research require HackerRank?

No current official page names HackerRank or another vendor for the 2027 engineering skills test. Older anonymous reports mention HackerRank, but they describe 2020-2021 processes and conflict on task count and timing.

Is there a mathematics test for software engineering interns?

G-Research's public engineering process names an online skills test, not a separate mathematics test. Quantitative reasoning still helps because the engineering teams build systems supporting research; prepare practical probability, statistics, numerical safety, and complexity reasoning without claiming a separate math round.

How long does the process take?

G-Research says the general engineering internship process typically takes one to two weeks across the three stages. Scheduling and team needs can vary, so treat that as a typical estimate rather than a guarantee.

Which programming language should I use?

Use the language in which you can write correct, testable code quickly and explain its standard data structures. If the invitation specifies a language or environment, follow it. If it does not, ask before the assessment instead of inferring a preferred language from a historical report.

Sources and further reading

Candidate and third-party reports are dated and may describe a different team or cycle; the application invitation remains authoritative.