A global trading firm is establishing a new core AI team in New York. It is their newest research group: a team that builds its own machine learning strategies and trades them.
It is a trading team first. Researchers, ML engineers and software engineers work side by side and own the full lifecycle, from the first experiment through production to revenue. A model that works in research goes live with the people who built it. Success is measured in P&L.
The research is model building. You will design and train new models, including LLMs and other foundation models, on large financial datasets that span high-frequency time series and unstructured text. The problems are sequential and noisy, and the papers you start from were rarely written with them in mind.
The work runs from transformers and state-space models to reinforcement learning, self-supervised learning and the GPU engineering needed to train at scale. You will also build their research tooling, including agentic workflows that run parts of the experiment loop. Every idea faces baselines, ablations and out-of-sample tests before it reaches a live market.
The Role
Design and train new models, including LLMs and other foundation models, for live trading systems
Develop ML methods for hard sequential and quantitative modelling problems
Run experiments with proper baselines, ablations and out-of-sample tests, and find out why an idea failed as well as whether it did
Turn papers into working code and extend them to domains they were never written for
Search large financial datasets, both time series and unstructured, for predictive signal
Train and evaluate at scale across GPU clusters
Build research tooling, including agentic workflows that automate parts of the experiment loop
Stack:
Python: PyTorch, JAX
GPU: CUDA, Triton, custom kernels
Distributed training across large GPU workloads
Architectures: transformers, efficient attention, SSMs, Mamba, RWKV
Required:
Grounding in machine learning, statistics, optimization and experimental design
A track record of independent ML research, from hypothesis through implementation to evaluation
Hands-on experience building and training models rather than integrating existing ones
Depth in modern architectures: transformers, foundation models, sequence models or state-space models
Strong Python, with PyTorch or JAX
Experience training and evaluating models on GPUs
Eligibility to work in the US
Pre Application:
Please do not apply if you’re looking for a contract or remote work.
You must be eligible to live and work in the US, without requiring sponsorship.
Please ensure you meet the required experience section prior to applying.
Allow 1-5 working days for a response to any job enquiry.
Your application is subject to our privacy policy (thurnpartners.com/privacy-policy).
Seniority Level
Mid-Senior level
Employment Type
Full-time