About the role
Join
the team redefining how the world experiences design. Hey, g'day, kia ora, 你好, hallo, vítejte! Thanks for stopping by. We know job hunting can be a little time consuming and you're probably keen to find out what's on offer, so we'll get straight to the point. Where and how you can work Our flagship Sydney campus is uniquely Canva - an extension of our Surry Hills neighbourhood. It’s a thoughtfully designed space with plenty of room to collaborate, focus, and connect. This role is based in Sydney, and we’re looking for someone who calls it home. Our hybrid way of working gives you the flexibility to work remotely, and to come together on campus for meaningful in-person collaboration and connection when it matters most. We trust our Canvanauts to choose the balance that empowers them and their team to achieve their goals. At Canva, our mission is to empower the world to design. To get cutting-edge research into the hands of millions of users faster, we're looking for a Machine Learning Engineer focused on research enablement and performance, turning promising experiments into stable, scalable, user-facing capabilities while making training and inference faster, cheaper, and more reliable.
About
the role: You'll be the bridge between research and production. Partnering closely with researchers, you'll ensure experimental code is production ready, integrate models into our monorepo, build shared libraries and services, and create the tooling and processes that let multiple model variants ship safely and quickly. You'll also work across the training stack, profiling and tuning PyTorch workloads, improving GPU utilisation, and shaping how we use distributed training and storage to get the most out of our compute. Your work shortens the research-to-user loop, reduces duplication, and ensures our ML features are reliable, observable, and easy for other teams to adopt. At the moment, this role is focused on: Research-to-Production Pipeline: Hardening experimental models (containerisation, tests, CI/CD), making them deployable for real users. Training Performance and GPU Efficiency: Profiling PyTorch training jobs, improving GPU utilisation, and applying techniques like mixed precision, efficient data loading, and distributed training strategies (FSDP, DDP, DeepSpeed) to reduce time and cost per experiment. Library development: Converting experiments into well-factored libraries with clear APIs, dependency hygiene, and versioning, so teams can import rather than copy-paste. Developer Experience & Documentation: Creating templates, examples, and guidance; offering supportive, high-signal communication so others can adopt libraries confidently. Reliability, Observability & Cost: Instrumenting services with metrics/logging/tracing, setting SLIs/SLOs, and optimising training and inference performance and spend. Primary
