PhD AI Research Intern: VLMs, RLHF & Multimodal Eval

black.ai; Greater London; Hybrid; GBP 22,000 - 29,000; Full time

Job summary

Canva is seeking current PhD students for a full-time, 16-week AI Research Internship in London to translate research into real-world AI-powered experiences. You will work with Canva's AI team on live, industry-scale projects, gaining hands-on data, production infrastructure, and deadlines while learning from researchers and engineers building Canva's next generation of AI.

The role focuses on designing and evaluating VLM-based methods, reward modeling, and combining RL with layered-generation

Qualifications

• You're a PhD student ready to bring research into product experience.
• Strong diffusion or flow-matching background with policy-gradient RL for generative models.
• Experience fine-tuning VLMs (LoRA) and designing prompts for evaluation tasks.
• Reward modelling, preference optimisation, pseudo-labelling and distillation.

Responsibilities

• Design and validate a rubric-guided per-layer VLM judge for RGBA layer decomposition.
• Build VLM-based methods for automatic evaluation of multi-layer designs.
• Turn VLM-based evaluators into reward functions to train generative models with RL.
• Distill judges into lightweight reward models scoring layered images from learned representations.
• Collaborate with research, engineering and product teams to move findings toward production.
• Contribute to the broader research community through publications.

Skills

PhD student

Diffusion models

Reinforcement learning

LoRA fine-tuning

Prompt design

Research publication

Education

PhD in progress

Tools

PyTorch

DeepSpeed

RL libraries

Job description

Canva is seeking current PhD students for a full-time, 16-week AI Research Internship in London to translate research into real-world AI-powered experiences. You will work with Canva's AI team on live, industry-scale projects, gaining hands-on data, production infrastructure, and deadlines while learning from researchers and engineers building Canva's next generation of AI.

The role focuses on designing and evaluating VLM-based methods, reward modeling, and combining RL with layered-generation