NEXIS identifies causal heterogeneous treatment effects by discovering Markov-blankets in pre-treatment data. It leverages multi-modal, multi-view measurements and scalable representations with minimal human input, enabling interpretable and actionable policy insights from controlled experiments.