• Build a learned world model of the welding process that predicts future system behavior under robot actions. • Develop multimodal neural simulators incorporating signals such as 3D scans, video, thermal data, and electrical measurements. • Design, train, and evaluate large-scale generative or dynamics models (e.g., video prediction, latent world models, 3D or spatiotemporal representations) capable of long-horizon rollouts. • Collaborate with reinforcement learning engineers by integrating the neural simulator into RL pipelines for policy training and evaluation. • Run research tracks in parallel with production development, including hypothesis-driven experimentation and ablation. • Partner closely with data and MLOps teams to support scalable training, evaluation, and deployment - while remaining comfortable owning pieces of the stack when needed. • Translate research prototypes into robust, maintainable production code when they prove valuable. • Validate simulator performance against real-world robotic welding data and support sim-to-real transfer.