Frontier AI Systems Research Engineer

Huawei Canada; Markham; On-site; CAD 127,000 - 225,000; Full time

Job summary

Huawei Canada is inviting a Researcher for a 12-month contract focusing on Frontier AI Technology Research. The role analyzes state-of-the-art AI model architectures and hardware workloads, developing mapping frameworks from AI models to hardware needs and deployment strategies.

You will evaluate performance trade-offs, contributing to next-gen AI accelerators with Roofline-based analyses and experimental kernel optimizations.

Qualifications

• Strong understanding of modern AI model architectures and emerging trends such as MoE and sparse/linear attention.
• Solid knowledge of AI accelerator architectures, memory hierarchies, interconnects and performance bottlenecks.
• Hands-on experience with AI kernel development using Triton/CUDA; familiarity with FlashAttention and kernel optimizations.
• Experience with LLM inference systems and serving frameworks; familiarity with PyTorch/JAX internals.
• Experience with hardware performance analysis and Roofline-based methodologies.

Responsibilities

• Track evolution of state-of-the-art AI model architectures and workloads for training and inference.
• Develop a framework mapping AI workloads to hardware requirements and deployment strategies.
• Evaluate architectural innovations and provide quantitative performance recommendations for AI accelerators.

Skills

MoE architectures

GPU/TPU hardware

AI kernel development

LLM inference optimization

Roofline analysis

Education

PhD in AI/CS or closely related field

Tools

Triton

TileLang

CUDA

PyTorch/JAX

Job description

Huawei Canada is inviting a Researcher for a 12-month contract focusing on Frontier AI Technology Research. The role analyzes state-of-the-art AI model architectures and hardware workloads, developing mapping frameworks from AI models to hardware needs and deployment strategies.

You will evaluate performance trade-offs, contributing to next-gen AI accelerators with Roofline-based analyses and experimental kernel optimizations.