Lan Feng is a PhD candidate at EPFL’s VITA Lab, advised by Prof. Alexandre Alahi, and currently a Research Intern at NVIDIA Research. His doctoral work develops data-centric methods for foundation models, spanning LLM agents, world models, and end-to-end driving, with a focus on principled data selection and closed-loop evaluation. In an earlier chapter, he studied Navigation Engineering at Wuhan University before turning to Robotics at ETH Zurich, modelling everything from human-to-robot handovers to realistic urban traffic. In his research, he is driven by the conviction that machine learning is increasingly limited by data rather than models or compute, and that the right subset of data matters far more than simply collecting more of it. This has led him to his current focus on data-centric pipelines and generative world models for autonomous driving, where the rare, long-tail scenarios that matter most push the limits of how machine learning captures the real world and human driving behaviour.