Machine Learning Engineer - GPU Performance & Low-Latency Inference

Apply Now
Job Description

We’re partnering with a leading global trading firm to hire a Machine Learning Engineer to make ML models run as fast as possible. You’ll work close to the hardware, squeezing latency out of real-time inference and speeding up training, in an environment where microseconds matter.


What you’ll do

  • Build and optimize low-latency inference pipelines that deliver real-time predictions in production
  • Profile and optimize ML workloads at the device level: memory, compute, kernels and compilers
  • Write and tune custom GPU code (CUDA, Triton or equivalent) where libraries fall short
  • Apply and validate low-precision techniques (FP8/INT8, quantization), knowing which optimizations stay bit-exact,
  • Optimize batching, KV caching and serving paths for latency-critical workloads.
  • Extend the internals of PyTorch and inference frameworks to raise performance on specific workloads


What you’ll bring

  • 3+ years in machine learning, focused on inference or GPU performance
  • Hands-on experience optimizing ML at the GPU level: CUDA, Triton, C++, TensorRT or similar
  • Experience with profiling tools (Nsight, torch.profiler) and reasoning about memory bandwidth, latency and throughput
  • A strong grasp of modern ML architectures, attention in particular, and how they map onto hardware
  • Production-quality Python (C++ a strong plus), with testing discipline and ownership of the systems you ship
  • A habit of measuring before optimizing, and the judgment to know when to build versus buy


Nice to have

  • Experience building real-time, low-latency prediction systems in high-performance environments
  • Experience with numerics: mixed precision, quantization and correctness at scale
  • Contributions to open-source projects such as PyTorch, vLLM, Triton or FlashAttention


No trading experience is required. What matters is deep, hands-on performance engineering and a drive to make things faster.

Apply Now