Quantitative ML Researcher – From Hypotheses to Production

Referment; New York (NY); On-site; USD 140,000 - 210,000; Full time

Job summary

Referment is seeking a machine learning researcher to tackle complex quantitative problems in systematic investing. The role combines hypothesis-driven experimentation with scalable implementation and real-world deployment.

You will merge deep ML expertise with empirical judgment, exploring new architectures and training methods while addressing computational bottlenecks for large-scale research.

Qualifications

• A bachelor's degree or higher and deep expertise in machine learning.
• Strong mathematical and statistical foundations, with excellent experimental judgment.
• Excellent programming skills and the ability to take research from concept to scalable implementation.
• Practical experience with PyTorch or JAX; CUDA, XLA, Triton, large-scale training or performance optimisation would be valuable.
• Intellectual curiosity and an interest in applying ML advances to difficult quantitative problems.

Responsibilities

• Develop and evaluate machine learning approaches for challenging quantitative research problems.
• Form hypotheses and design rigorous experiments around predictive performance, robustness and generalisation.
• Explore new model architectures, representations, objectives and learning techniques.
• Prototype and benchmark ideas using large financial and other complex datasets.
• Improve training and implementation efficiency, including GPU optimisation where useful.
• Work with quantitative researchers and developers to move successful research towards application.

Skills

Machine learning

Mathematics

Programming

Education

Bachelor's degree

Tools

PyTorch

JAX

CUDA

XLA

Triton

Job description

Referment is seeking a machine learning researcher to tackle complex quantitative problems in systematic investing. The role combines hypothesis-driven experimentation with scalable implementation and real-world deployment.

You will merge deep ML expertise with empirical judgment, exploring new architectures and training methods while addressing computational bottlenecks for large-scale research.