Deep Learning Engineer, World Models

📋 Description

• Design and train multimodal world models (video/state/action/language) with diffusion/transformers.
• Build action-conditioned video prediction and dynamics with long-horizon physical consistency.
• Develop learned-simulator evaluation: offline scoring, predict real-world success pre-deploy.
• Generate synthetic rollouts and counterfactuals to augment policy training.
• Establish fidelity metrics and calibration for trust vs reality.
• Build data pipelines from telemetry, logs, and video into training corpora.

🎯 Requirements

• Track record of training large generative models — video/multimodal — with shipped models.
• Deep hands-on experience with diffusion models, autoregressive transformers, latent-variable models, or video prediction.
• Experience with large-scale distributed training: streaming data, checkpointing, debugging numerics.
• Strong Python + PyTorch/JAX; profile kernels, optimize data loaders, maintainable code.
• Empirical rigor: careful evaluations, honest baselines, and clear experiments.
• Excitement about grounding generative models in physical reality rather than pixels alone.

🎁 Benefits

• Competitive equity: stock options with meaningful upside.
• 30+ paid days off, including 23 days annual leave and UK holidays.
• Private healthcare, including virtual and in-person care.
• Pension scheme with 8% total contribution (5% employee, 3% employer).
• Free breakfast, catered lunch, and snacks in-office.
• Real ownership and direct impact on product direction from day one.