ML Performance Engineer: Production GPU Optimizer

Goliath Partners Inc.; New York (NY); On-site; USD 200,000 - 300,000; Full time

Job summary

Goliath Partners Inc. in New York is seeking a Machine Learning Performance Engineer to optimize production inference and accelerate models with custom GPU kernels.

You will tackle bottlenecks, memory behavior, and mixed-precision execution while aiming for the fastest, most reliable deployments in a fast-paced trading environment. Based in New York, you will collaborate with ML researchers, computing infrastructure, and hardware accelerators to shape model deployment strategies and ensure

Qualifications

• 2+ years improving deep learning inference performance in production environments where speed or throughput is critical.
• Strong Python and C++, with experience working beneath the high-level APIs of frameworks such as PyTorch or JAX.
• Practical experience developing custom GPU kernels and using inference optimization tools.
• Detailed knowledge of GPU execution, memory behavior, and mixed-precision computation.
• A measurement-driven approach to profiling systems and comparing performance across different hardware architectures.

Responsibilities

• Write and tune GPU kernels to accelerate production inference.
• Investigate performance bottlenecks, from memory access patterns to communication between hardware components.
• Reduce model size and computational demands while preserving numerical stability.
• Compare hardware options and help determine the best deployment approach for demanding trading workloads.
• Partner with researchers and engineering teams to bring optimized models into production.

Skills

Python

C++

PyTorch

JAX

GPU kernels

Performance profiling

Hardware acceleration

Inference optimization

GPU memory

Low latency

FPGA/ASIC experience

Measurement-driven profiling

Job description

Goliath Partners Inc. in New York is seeking a Machine Learning Performance Engineer to optimize production inference and accelerate models with custom GPU kernels.

You will tackle bottlenecks, memory behavior, and mixed-precision execution while aiming for the fastest, most reliable deployments in a fast-paced trading environment. Based in New York, you will collaborate with ML researchers, computing infrastructure, and hardware accelerators to shape model deployment strategies and ensure