Headlands Technologies Interview Guide

The camera stays on and screenshots are blocked. Paper is used for calculations. Candidates describe this format consistently from April 2025 to October 2026, and one counted 18 questions. Negative marking, the pass bar, the sitting window and retakes are not reported. If you have sat it, tell us what you saw.

Reported questions

• Multiple choice. What cross-validation reduces, with bias, variance and the 2 error types as options. The assumptions behind the F-test, t-test and chi-square test. Correlation, covariance and basic probability. The statistics is demanding, and the terminology needs preparation.
• Worked problems. In a regression of y on x1, how the t-statistic of the x1 coefficient changes when you add a regressor orthogonal to x1. A strategy backtests at a Sharpe ratio of 8 and then makes nothing over k live days: at what k should you doubt it? The estimator c that minimises the expected absolute deviation of a sample. A third Sharpe question involving several strategies.
• C++ numerical. The mean over a sliding window of fixed size on an array of floats, with correct handling of infinity, negative infinity and NaN. Know how C++ represents these values before the test.
• Algorithms. The minimum number of character changes so that no 2 adjacent characters in a string are equal. A LeetCode easy with a medium follow-up.

The engineering test

Software engineers still report the older format: 1 problem in 90 to 120 minutes, C++ only. The classic is a job runner: read a CSV of job id, runtime and next job id from standard input, check that the input forms valid chains, and output totals, counts and the average time per chain. Only 2 sample tests are given, the hidden tests are much stricter, and the traps are cycles, chains over 86,400 seconds and malformed rows. A newer variant uses file and timestamp records with union-find. Answer in C++, as the test asks.

How to prepare

• Vocabulary first. Bias against variance, the 2 error types, the assumptions of each standard test. The hypothesis testing and cross-validation lessons are the right level.
• OLS algebra. What adding an orthogonal regressor does to a t-statistic, from the multiple regression lesson. Why the median minimises absolute error.
• Sharpe arithmetic. How many days of zero return it takes to make a backtest's Sharpe ratio implausible. The reading a backtest section covers it.
• C++ numerics. std::isnan, std::isinf, how NaN propagates through a running sum, and the floating point lesson. The streaming statistics lesson covers the window.
• The engineering problem. Write your own tests for cycles, orphans and overflow before you submit.

A 60-minute video call, one on one with a researcher, on Zoom or Teams. Some candidates get 2. The first interview can be largely about your resume.

What they cover

• Regression and inference theory. The assumptions of linear regression and why each is needed. Why we minimise mean squared error. The theoretical basis of OLS. The difference between a z-test and a t-test. Two datasets, 10 samples and a million, both giving p = 0.04: which is more credible?
• Machine learning. Regularisation, principal component analysis, and whether you have heard of a U-Net.
• Applied judgement. Blend 2 strategies knowing only their means and distributions. A survey correlates IQ with work motivation: what can you conclude?
• Numerical methods. Optimising a numerical linear algebra algorithm.
• Your projects. Machine learning and data cleaning techniques from your own work.

Expect mostly statistics, machine learning and C++, with few brainteasers. Another found them very random and advised the Green Book plus machine learning and deep learning.

How to prepare

State the OLS assumptions and the reason for each. Work p-values and statistical against practical significance for the sample size question, then ridge and lasso and the unsupervised learning section for regularisation and PCA. Keep the probability questions warm for the few brainteasers.

In person at the office, all on the same day. Candidates report 6 rounds for a developer, and 8 for an experienced quant after 2 phone screens. The onsite can include 2 live HackerRank rounds where you compile on the spot, run test cases and need the standard library, and all of the firm's quant coding is in C++. No report describes a whiteboard, session lengths, breaks or lunch, and no game, case or take-home is reported.

Statistics, OLS, hypothesis testing and machine learning dominate every reported technical round, with background and project questions throughout. Software engineers report both LeetCode-style problems and C++ trivia, and later rounds on algorithms, operating systems, concurrency and open-ended systems questions, including how you would find and remove unused code from a codebase.

The interviewer style is the thing candidates mention most. Candidates describe an interviewer who asks one technical question and wants to hear the specific answer he has in mind. Others describe interviewers who give few hints. Because Headlands is small, the round depends on what your interviewer wants to talk about. Some interviewers walk you through what you missed, so the feel varies by person.

How to prepare

Write C++ with the standard library every day for 2 weeks before the onsite, compiling and running against your own tests. The Quant Dev 50 and the problems below are the right material. Prepare your projects to the level of the equations and the data cleaning decisions, because they come up in every round.

The C++ HackerRank above, then a one-on-one Zoom screen with a developer or researcher, then the onsite. Reported screen problems: a queue with push, pop and get-mean; merging several data streams in C++ with a multithreaded non-blocking follow-up; order interval overlap and cancellation; the sum of the first n integers without the formula, then compile-time and const-correct variants, then a custom hash map for ticker lookups that cuts collisions and memory; and a trading-flavoured object design question for junior engineers. The systems programming course covers the internals and concurrency these rounds reach for.