Machine Learning Researcher Internship, June-September

Jane Street’s Machine Learning Researcher Internship is built around the same kind of work its full-time ML researchers do. Interns work alongside experienced researchers on projects chosen because they combine interesting machine-learning questions with direct relevance to real systematic-trading strategies.

What you will work on

Most of the internship is spent on research projects drawn from the work of Jane Street’s ML team. Depending on the project, this can include:

• Running an end-to-end study on a dataset that has not yet been fully explored.
• Testing a new modelling approach on a difficult trading problem.
• Exploring more open-ended approaches when the team does not yet know the best solution.
• Working directly with market data.
• Tuning hyperparameters and comparing modelling choices.
• Debugging training problems and understanding why a model is not behaving as expected.
• Analysing model predictions and deciding what the results suggest should be tested next.

The problems are intentionally research-heavy and often do not have a single clean answer. Interns may need input from researchers, traders, and engineers with different areas of expertise.

Data, models, and compute

Jane Street’s research, technology, and trading teams work closely together. ML researchers have access to petabytes of data, a CPU cluster with hundreds of thousands of cores, and a growing GPU cluster with tens of thousands of high-end GPUs.

The trading setting creates specific ML challenges, including large models, non-stationary datasets, and a competitive multi-agent environment. Interns use established ML methods as well as newer techniques when appropriate, and the programme includes classes and activities explaining how Jane Street approaches markets and trains practical models.

Because the work is highly proprietary, Jane Street states that research produced during the internship is unlikely to be suitable for external academic publication.

Interview process

The Machine Learning Research interview process begins with several Zoom interviews and then moves to in-person interviews at the office being applied to.

The ML-focused interviews test how candidates reason through difficult deep-learning problems rather than whether they can repeat standard textbook answers. Candidates may be asked to adapt modelling techniques to unusual settings, work under unexpected constraints, identify subtle bugs, and reason from first principles.

Ideas are expected to become concrete. Candidates write code, implement proposed approaches, and then explain how they would change the model or experiment after seeing the results. Jane Street mainly uses PyTorch internally, but candidates can use another deep-learning framework if they are more comfortable with it. Exact syntax and argument names do not need to be memorised and may be looked up during the interview.

The broader process can also include:

• Coding: software-engineering-style questions with less emphasis on clever algorithms and more emphasis on clear, well-organised code.
• Data analysis and exploration: questions around data quality, feature construction, and statistical reasoning.
• Research discussion: a detailed conversation about previous research experience from either academia or industry.

Throughout the process, Jane Street focuses heavily on reasoning. Interviewers may ask why a particular model, loss function, or hyperparameter was chosen, which hypothesis should be tested next, and how experimental results should change the next step. Candidates with experience in a particular area may also be asked to apply that knowledge to relevant modelling problems or discuss recent architectures and optimisation ideas.

What Jane Street looks for

• An undergraduate, PhD student, or postdoc with practical experience working on machine-learning problems.
• Strong logical and mathematical reasoning across unfamiliar problems.
• Interest in the wider machine-learning landscape and in applying state-of-the-art techniques from different problem domains.
• Comfort working with a broad set of models and practical modelling techniques.
• The ability to implement and iterate quickly in Python using a preferred ML framework.
• Willingness to ask questions, acknowledge mistakes, and learn quickly.
• Fluency in English.

How to prepare

Jane Street recommends reviewing standard deep-learning models and techniques and practising the process of training, evaluating, and improving models in code. For the coding portion, the emphasis is on writing clear and structured code rather than solving unusually tricky algorithmic puzzles.