Source and shape portfolio use cases: interface with Flagship teams and portfolio company end-users to identify use cases, source data assets, define success criteria, and create end-user feedback loops for reasoning engines that are scientifically meaningful, tractable, and answer high-value questions with real world application. Guide technical architecture through a biological lens: guide multi-modal, multi-scale model design and ensure the system can ingest and reason over modalities relevant to mechanism-of-action reasoning, target discovery, perturbation biology, pathway reasoning, protein function, and autonomous discovery workflows.