Senior Machine Learning Systems Engineer (Training Optimization)

Company Description: About the Group/Team We're the CORE team within the Generative AI supergroup. Our mission is to invent foundational technologies that will power the future of AI-assisted design. From large-scale models to groundbreaking research, our team builds the technical core of Canva’s creative intelligence engine. We collaborate globally to ship research that makes a real impact—from smart editing to AI video tools—at massive scale. Job Description: About the Role/Specialty As a Senior Machine Learning Systems Engineer, you’ll lead efforts to scale and optimize the training system for our large-scale multimodal and foundation models. You’ll design distributed training systems using Megatron-LM, NVIDIA NeMo, FSDP, and Triton—pushing the limits of performance across compute, memory, and communication layers. You'll sit at the intersection of systems and AI research, directly shaping how we train the models that will power Canva’s next generation of products. What you’ll do (responsibilities) You’ll design, implement, and optimize large-scale machine learning systems for training You’ll improve all aspects of performance, including GPU utilization, communication overhead, and memory efficiency. You’ll partner with research and modeling teams to align systems with algorithmic needs. You’ll evaluate and apply best practices for distributed training using industry-leading frameworks. You’ll dive deep into low-level optimization, including custom CUDA or Triton kernels. You’ll debug, profile, and fine-tune training workflows to unlock new levels of scalability. Qualifications: What we're looking for We’re looking for a systems-first engineer who thrives in fast-paced, high-impact environments. You’re deeply familiar with distributed model training at scale and understand the nuances of optimizing compute at every level of the stack. You're excited by challenges that stretch current boundaries, and you’re a strong collaborator who communicates clearly across domains. Strong background in LLMs, multimodal AI, or diffusion models. Proficiency in Python. Familiarity with a system programming language (e.g. C++ or Rust) is a plus. Deep knowledge of PyTorch or JAX as well as libraries such as Megatron-LM, NeMo, or DeepSpeed. Familiarity with common optimization techniques such as FSDP/ZeRO, gradient checkpointing, or low-precision data types. Hands-on experience writing custom GPU kernels in CUDA or Triton. Excellent communication and problem-solving skills, incl. full proficiency in English.