Research Engineer, Pre-Training

Responsibilities

• Develop massive-scale foundation models and fault-tolerant training infrastructure scaling across thousands of GPUs and TPUs
• Own and drive the entire training stack, including building data pipelines that stream terabytes per second
• Design custom kernels to unlock significant efficiency gains
• Co-design novel architectures with researchers and pioneer cutting-edge approaches to mixed-precision training and model parallelism
• Balance ambitious research goals with practical engineering constraints to impact live trading

Requirements

• Expertise in large-scale distributed training (MFU, throughput, convergence, cost-per-token)
• Published research in ML systems, architectures, scaling laws, or efficient training methods
• Background in HPC, numerical computing, or distributed systems (GPUs/TPUs, NVLink/InfiniBand, Kubernetes/Slurm)
• Proficiency in Python and deep experience with PyTorch or JAX
• Advanced degree (MS or PhD) in Computer Science, Machine Learning, Physics, Mathematics, or a related quantitative field

Preferred Qualifications

• Expertise in CUDA kernel development, Triton/Pallas/CuTe DSLs, or XLA optimization
• Experience with hardware acceleration such as FPGA or ASIC
• Knowledge of reinforcement learning, post-training, or fine-tuning techniques
• Knowledge of financial markets or trading

Benefits

• Discretionary bonus eligibility
• Medical, dental, and vision insurance
• HSA, FSA, and Dependent Care options
• Retirement plan with employer match
• Paid vacation, holidays, and parental leave
• Wellness Programs

About the Company

Jump Trading Group is committed to world-class research, empowering exceptional talents in Mathematics, Physics, and Computer Science to push scientific boundaries and apply cutting-edge research to global financial markets. We design, develop, and deploy technologies that change our world and partner with leading global research organizations to solve complex problems.