Medicine today is imprecise. For the top 20 drugs in the US, 80% of patients are non-responders. The dream of precision health is to develop a data-driven, learning system where new health information is instantly incorporated to optimize care delivery and accelerate biomedical discovery. In reality, however, the health ecosystem is plagued by an overwhelming amount of unstructured data and unscalable manual processing. My life-long aspiration is developing next-generation AI technology to accelerate progress in access, safety, and preventative care for precision health. At Microsoft, I lead biomedical AI research and incubation, with a particular focus on scaling real-world evidence generation by structuring all medical data. We're among the first to study the development and applications of large language models in biomedicine. To incorporate additional information-rich modalities such as medical images and omics, we are developing multi-modal learning and fusion methods for predicting disease progression and drug response. We're also incorporating advanced causal learning methods to combat confounders in population-level observational data. Through deep collaboration with key stakeholders in healthcare and life sciences, we have made promising progress in end-to-end real-world applications.