AI Performance Engineer: CUDA & Multi-GPU Optimization

Brillfy Technology Inc; United States; On-site; USD 150,000 - 210,000; Full time

Job summary

Brillfy Technology Inc. is seeking a hands-on GPU/AI systems engineer focused on CUDA optimization, multi-GPU performance, and system-wide profiling.

The role requires deep expertise in GPU architecture and performance tuning, with experience across NVIDIA tools and distributed AI stacks. The ideal candidate will profile AI/ML workloads, identify bottlenecks, and implement latency and throughput enhancements across compute, memory, and networking layers, collaborating with AI, DevOps, and

Qualifications

• Strong hands-on CUDA programming and GPU performance optimization.
• Deep understanding of GPU architecture and memory hierarchy.
• Experience with Nsight and CUDA profiling tools.

Responsibilities

• Profile and optimize AI/ML workloads across multi-GPU and multi-node systems.
• Identify bottlenecks across compute, memory, networking, and orchestration layers.
• Optimize CUDA kernels (memory coalescing, shared memory usage, occupancy tuning).
• Improve inference performance using TensorRT, Triton, DeepStream, NeMo.
• Analyze and improve latency, throughput, GPU utilization, and memory efficiency.
• Work on distributed AI systems using Apache Ray, NCCL, Kubernetes GPU scheduling.
• Build benchmarking frameworks and performance monitoring systems.
• Collaborate with AI, DevOps, and Infrastructure teams for system-wide optimization.

Skills

CUDA programming

GPU architecture

Performance profiling

NVIDIA ecosystem (TensorRT, Triton, Ne

Tools

Nsight tools

Job description

Brillfy Technology Inc. is seeking a hands-on GPU/AI systems engineer focused on CUDA optimization, multi-GPU performance, and system-wide profiling.

The role requires deep expertise in GPU architecture and performance tuning, with experience across NVIDIA tools and distributed AI stacks. The ideal candidate will profile AI/ML workloads, identify bottlenecks, and implement latency and throughput enhancements across compute, memory, and networking layers, collaborating with AI, DevOps, and