Machine Learning Engineer

About the Role

Research at Uncharted Network moves fast, and the ML Engineer is the reason production keeps up. You will own the infrastructure that takes models from the research bench to intraday retraining cycles — building training pipelines, GPU orchestration layers, and low-latency inference servers that deliver predictions to live strategies with consistent reliability. This is not a support role: you will work alongside ML Researchers to co-design the systems they depend on, push the state of the platform, and own every component end to end. Reliability, throughput, and reproducibility are the job.

Responsibilities

• Design and operate distributed training pipelines for financial deep learning models, including multi-GPU and multi-node jobs
• Build and maintain low-latency inference infrastructure serving model predictions to the live trading engine
• Implement GPU orchestration, job scheduling, and resource management across research and production clusters
• Tighten feedback cycles between research iteration and production deployment through platform improvements and tooling
• Monitor training runs, model performance drift, and inference latency in production environments
• Collaborate with quantitative researchers to co-design experiment frameworks and evaluation tooling

Requirements

• 4+ years of experience building and operating ML training and inference infrastructure end to end
• Strong working knowledge of PyTorch and/or JAX; experience with CUDA or Triton for custom kernels is a significant advantage
• Experience with distributed training frameworks (NCCL, DeepSpeed, or equivalent)
• Proficiency in Python and at least one system-level language (Rust, Go, or C++)
• Understanding of ML model lifecycle management: versioning, deployment, monitoring, and rollback
• Strong mathematical intuition for debugging model performance and training instability

Nice to Have

• Low-level GPU profiling experience using NSight Systems or NSight Compute
• Knowledge of CUTLASS, CUB, or Triton for custom CUDA kernel development
• Experience with high-throughput inference serving (Triton Inference Server, vLLM, or custom solutions)