Principal Architect – Hardware Efficient AI Foundation Model Training

About the role

Huawei Canada’s Computing Data Application Acceleration Lab is seeking a Principal Architect to lead hardware‑efficient AI foundation model training. The role focuses on full‑stack innovations, software‑hardware co‑design, and optimizing data efficiency across storage and runtime layers.

Key responsibilities

• Collaborate with internal and external partners to design foundational model architectures for LLM, code, and multimodal sub‑fields, driving breakthroughs in post‑training and continual training.
• Define technical requirements for large‑scale distributed training and inference infrastructures, including parallelization strategies and operator fusion.
• Analyze computational characteristics of emerging AI architectures to ensure accuracy, performance, and hardware evolution.

Required profile

• Proven experience training and optimizing cutting‑edge AI models at scale (10B+ parameters).
• Deep knowledge of modern AI architectures such as long‑sequence models, reinforcement learning, multimodal systems, and autonomous agents.
• Strong understanding of AI algorithm mechanisms and their implementation.
• Hands‑on expertise with AI frameworks (e.g., PyTorch, vLLM, SGLang) and mainstream distributed training/inference techniques.
• Familiarity with AI chip architectures (GPU, NPU, TPU) and memory hierarchy/interconnect technologies.
• PhD in AI architecture, computer architecture, or a related field is preferred.
• Solid publication record in AI systems or chip design is an asset.

Required skills

• Training and optimizing large AI models (10B+ parameters).
• AI architectures: long‑sequence, reinforcement learning, multimodal, agents.
• AI frameworks: PyTorch, vLLM, SGLang.
• Distributed training and inference techniques.
• AI chip architectures: GPU, NPU, TPU.
• Memory hierarchy and interconnect technologies.