Principal ML Infrastructure Engineer

Overview

AI Infrastructure Engineer (GPU Systems & Model Deployment) (Principal and Entry level available)

We are seeking an AI Infrastructure Engineer to design and optimize high-performance systems that enable machine learning models to run reliably and efficiently in production environments. This role is focused on GPU-accelerated inference, low-latency model serving, and bridging the gap between research models and real-world deployment. You will work closely with ML researchers and software engineers to ensure models are production-ready, scalable, and performant.

This is a hands-on systems role with a strong emphasis on C++, CUDA, and GPU inference optimisation
.

• Design and maintain GPU-accelerated infrastructure for deploying machine learning models in production
• Build and optimize high-throughput, low-latency inference pipelines
• Develop and maintain performance-critical components in C++
• Optimize GPU utilization through CUDA programming and kernel tuning
• Support model conversion, optimization, and deployment using inference runtimes
• Partner with ML researchers to transition models from experimentation to production
• Diagnose and improve system performance relative to baseline benchmarks
• Ensure deployed systems are reliable, observable, and maintainable in production environments

• Masters or PhD required
• Strong C++ expertise with experience writing and optimizing production-grade systems
• Hands-on CUDA programming experience and GPU performance optimization
• Solid understanding of GPU architectures and memory management

• Experience with TensorRT or similar GPU inference runtimes
• 1–7 years of experience as a Software Development Engineer supporting production model deployment
• Experience with model optimization, quantization, or runtime acceleration techniques
• Exposure to ML frameworks (e.g., PyTorch, Tensor Flow) from a systems or deployment perspective
• Experience working with containerized environments and CI/CD pipelines

• C++, CUDA
• GPU inference runtimes (e.g., Tensor
  
  RT)
• Linux, containers, cloud or on-prem GPU systems