Machine Leaning Performance Engineer (Inference)

About this role

Tower Research Capital seeks a Machine Learning Performance Engineer to optimize inference pipelines for microsecond-level latency in quantitative trading systems. You will architect and optimize ML inference across CPUs, GPUs, and FPGAs, collaborating with research and infrastructure teams to push hardware performance to its limits.

What you'll do

• Evaluate and benchmark diverse inference platforms (CPUs, GPUs, FPGAs) to guide infrastructure deployment decisions
• Optimize execution across memory hierarchies and resolve bottlenecks in the end-to-end inference lifecycle
• Develop highly optimized GPU kernels and integrate specialized performance libraries for maximum computational throughput
• Implement model optimization techniques (quantization, pruning, distillation) for compact, low-latency event-level inference
• Collaborate with ML Researchers, HPC Engineers, FPGA Engineers, and Datacenter Engineers on deployment strategies
• Design inference strategies that fit within thermal, power, and operational hardware constraints

What they're looking for

• Deep learning inference optimization in latency-sensitive production environments
• GPU kernel development and CUDA programming
• ML frameworks: PyTorch, JAX, with strong Python and C++ expertise
• Mixed-precision computation and quantization techniques
• Performance profiling tools (Nsight Systems, Nsight Compute)
• Optimization libraries and compilers (Triton, TensorRT, ONNX, IREE, cuBLAS, CUTLASS)
• GPU microarchitecture knowledge (SM execution, warp scheduling, memory hierarchy)
• Cross-architecture benchmarking and performance analysis

Benefits

• Competitive annual base salary of $200,000–$300,000
• Eligible for discretionary bonus
• Collaborative, ego-free work culture
• Open concept workplace with casual dress code
• Well-stocked office kitchens
• Global organization with 25+ year track record of innovation