Machine Learning Performance Engineer

Machine Learning Performance Engineer

A global quantitative trading firm is seeking a Machine Learning Performance Engineer to improve the speed, efficiency, and reliability of the systems powering its ML strategies.

You’ll work alongside quantitative researchers, hardware specialists, and software engineers to optimize demanding training and inference workloads. Your work will directly influence how quickly models can be evaluated and how effectively they operate in live trading.

What you’ll do

\n
• Optimize GPU workloads across training and inference, addressing bottlenecks in memory usage, computation, and latency.
\n
• Build and strengthen low-latency inference pipelines that deliver reliable, real-time predictions.
\n
• Improve performance across kernels, compilers, acceleration libraries, and ML framework internals.
\n
• Develop scalable model-serving infrastructure for high-volume trading data.
\n
• Evaluate optimization techniques while maintaining numerical correctness and model quality.
\n
• Assess and integrate tools that improve performance across the ML lifecycle.
\n
• Collaborate with researchers and engineers to measure results and carry improvements into production.
\n

What you’ll bring

\n
• 3+ years of relevant ML experience, with a focus on inference systems, training performance, or GPU optimization.
\n
• Strong production software engineering skills, including Python, disciplined testing, and ownership of deployed systems.
\n
• Depth in one or more of: GPU performance engineering, low-level optimization, or production inference.
\n
• An understanding of device-level memory, compute, and latency, with experience using technologies such as CUDA, Triton, or equivalent tools.
\n
• The ability to reason about ML workloads beyond high-level APIs, including performance and correctness tradeoffs.
\n
• A measurement-driven approach to troubleshooting and sound judgment on building solutions versus adopting existing tools.
\n
• Clear communication and the ability to work through technical decisions collaboratively.
\n

Experience with reduced precision, KV caching, batching, and bit-exact optimizations is particularly relevant. Cloud platforms, orchestration tools, and open-source contributions are also valued.