Machine Learning Engineer

New Yorkfull timeOn Site

mid

$175k–$250k

H1B: no

Imc Machine Learning Engineer at IMC builds systems for training and deploying large-scale ML models, collaborating with researchers and engineers to optimize GPU acceleration and distributed computing. The role focuses on either ML Research Engineering or ML Performance Engineering to improve training pipelines or inference performance.

Responsibilities

Develop large-scale distributed training pipelines that manage complex datasets and models.
Partner with quantitative researchers to automate ML experiments, hyperparameter tuning and model retraining.
Work with HPC specialists to optimise workflows, improve training speed and reduce compute cost.
Build and optimise low-latency inference pipelines, ensuring models deliver real-time predictions in production systems.
Maximise performance in training and inference using GPU hardware and acceleration libraries.
Extend and improve the internals of machine learning frameworks and open-source tools to raise their performance for our workloads.
Design scalable model infrastructure capable of handling high-volume trading data and delivering real-time, high-accuracy predictions.
Evaluate and roll out third-party tools that enhance model development, training and inference capabilities.
3+ years of experience in machine learning, with a focus on training or inference systems.

Qualification

Required

Production software engineering — Python, testing discipline and genuine ownership of systems once they ship.
Machine learning fundamentals — you understand modern ML workloads beyond high-level APIs and can reason about model behaviour, correctness and performance at the level relevant to your area of expertise.
A systems mindset — you reach for measurement before opinion on performance, scale or reliability issues, and can make sound build-versus-buy decisions.
Strong collaboration — comfortable explaining the reasoning behind your technical decisions, working through differing perspectives with researchers and adapting your approach when the evidence supports it.
Distributed training and large-scale systems — parallelism strategy, data pipelines and checkpointing; frameworks such as PyTorch, TensorFlow or JAX.
GPU performance and low-level optimisation — reasoning about memory, compute and latency at the device level, and optimising performance across kernels, compilers and hardware (e.g. CUDA, Triton or equivalent).
Production inference systems — hands-on experience building and hardening real-time, low-latency prediction pipelines in high-performance environments.
Numerics and correctness at scale — low-precision effects, KV caching, batching, and knowing which optimisations stay bit-exact.
Exposure to cloud platforms and orchestration tools.
A track record of contributing to open-source projects in machine learning, data science or distributed systems.