Senior Research Scientist - Design Generation (Sydney)

About the Group/Team
Canva Research sits within Canva's Generative AI supergroup, home to Canva's original AI research. The Design Generation team's mission is to empower Canva to do AI magic in the design space, building the generative models and techniques that power AI-assisted design creation for hundreds of millions of users.
About the Role/Specialty
As a Senior Research Scientist on Design Generation, you'll work on the generative modelling problems behind how Canva creates and understands design content. That means training and evaluating design generation models at scale - from wrangling large-scale datasets, crafting training objectives that draw out the most from this data to improve design quality, evaluating them on criteria that users actually care about, and working with product teams to help ship those models to 250M+ users. You’ll take ownership of scoped research projects end-to-end, from framing an open problem through to a shipped or clearly validated result.
What you’ll do (responsibilities)
Train and evaluate generative models like LLMs for design generation
Design and run experiments to test hypotheses about model architecture, training approach, and data
Build and improve evaluation methods and benchmarks for measuring generative design quality
Translate research findings into techniques and approaches the team can move toward production
Investigate model quality issues, failure modes, and gaps against user and business needs
Stay current with generative modelling research and bring relevant advances into the team's work
Communicate research findings and trade-offs clearly, including to non-research stakeholders, and share knowledge across the team
What we're looking for:
You've got hands-on experience training generative models, ideally with exposure to image ordesigngeneration, not just applying pretrained models off the shelf. You're comfortable defining and moving the metrics that matter for generative quality, not just chasing benchmark numbers in isolation. You can take a scoped, often ambiguous research problem from framing to a validated result with minimal guidance, bringing your own judgement to how it should be approached. You're highly autonomous, comfortable with the ambiguity that comes with frontier research, and you want your work to show up in something real people use every day.
You've got hands-on, practical experience fine-tuning large language and multimodal models (LLMs/MLLMs) at scale in an industrial research setting.
You know the tricks of the trade: wrangling large-scale data pipelines, designing novel training objectives that extract the most from the data to improve design quality, and training and evaluating across multi-GPU setups
You've got a PhD in AI, machine learning, or a related field, plus at least two years of strong hands-on industry experience. We will also consider candidates without a PhD, but with at least 6 years of industry experience.
A background in diffusion or image generation is a bonus, not a requirement — the core need here is language and multimodal LLM depth
What You'll Learn & Build:
Deep experience researching and training generative models for design at Canva's scale
Exposure to how research, product, and infra decisions intersect across Canva Research and the wider Generative AI supergroup
The chance to work on frontier generative modelling problems as the space develops

Employment type
Full-time

Work arrangement
No