We are seeking an early-career research scientist.
You will take open, loosely defined problems from first question to a working, validated, defensible result. This is hands-on applied research: you will own the problem end to end, build the solution yourself, and be accountable for whether the evidence actually holds.
What you will do
- Take ownership of applied machine-learning research problems end to end: framing the question, assembling and cleaning the data, choosing and implementing the method, evaluating it honestly, and reaching a defensible conclusion.
- Build and validate models on messy, real-world data, with rigorous attention to data quality, data leakage, regime change, and genuine out-of-sample performance.
- Implement methods from first principles where needed, in Python and, where appropriate, a compiled or systems language.
- Design evaluation harnesses and benchmarks so that your own results, and the team's, are reproducible and falsifiable rather than merely asserted.
- Communicate your reasoning and results clearly to both technical and non-technical colleagues, and independently reproduce and pressure-test teammates' results.
- Work collaboratively in a small, high-density team, and raise the way the team works, not only your own output.
Required qualifications
An outstanding academic record in a highly quantitative discipline, for example computer science, machine learning, mathematics, statistics, physics, engineering, or a computational natural science, at first-class or distinction level or the equivalent. A master's degree, or near completion of one, is preferred.Demonstrated applied machine-learning research backed by a concrete artifact you can show and defend: a thesis, a paper, a benchmark, a shipped system, or a strong competition result.Strong programming ability, with Python essential, including real experience implementing algorithms from first principles rather than only calling existing libraries.Sound scientific method and data rigor: point-in-time and out-of-sample discipline, a working awareness of leakage and overfitting, and hands-on experience with imperfect, real-world data.Clear written and spoken communication of technical work, in English.Evidence of independent, self-directed problem-solving and genuine curiosity about hard, open-ended problems.
Preferred
Experience with modern deep-learning frameworks, for example PyTorch.Working experience with a compiled or systems language, for example C++.Depth in one or more of: causal inference, time-series modelling, inverse problems, Bayesian methods, numerical optimisation, or signal processing.Experience turning a research question into a reproducible benchmark or evaluation framework with known ground truth.A cross-disciplinary background that brings a methods toolkit the team would not otherwise have.Some applied or industry machine-learning experience.