Andy Beam

Andy Beam, also known as Andrew Beam, is Chief Technology Officer at Lila Sciences, where he leads development of AI for scientific discovery. His work spans machine learning, biomedicine and causal inference. Before Lila, he co-founded Generate:Biomedicines and helped build a generative AI platform for protein and therapeutic design.[1][2]

In his July 2026 Latent Space interview with Rafa Gómez-Bombarelli, Lila’s CSO for physical sciences, Beam explained the laboratory through computing analogies: instruments as nodes on a graph, a magnetically levitating transport layer as a PCI bus, and experiment orchestration as a Slurm queue. The analogy makes concrete the task of coordinating scientific equipment as computational infrastructure.[1]ⁿ¹

Beam argues that experimentally verified reasoning traces offer a kind of training data effectively absent from the internet. Lila treats reinforcement learning as a data-generation mechanism, with nature providing verification through experiments. Beam also emphasizes the runtime constraints of physical experiments: the ribosome cannot simply be made to go faster.[1]

In the joint interview, Beam and Gómez-Bombarelli discussed Lila’s choice to prioritize flexibility and generalizability over raw throughput, retaining human work where automation does not pay. Lila favors fast successive rounds of experiments over large, noisy multiplexed screens. Some measurements can nevertheless be accelerated substantially: Gómez-Bombarelli’s team rebuilt a gas-sorption measurement to run roughly 2,500 times faster.[1]

Before his work at Lila, Beam worked on protein engineering and medical AI. Those earlier roles provide context for his experience building models around biological and clinical data.[2][3]

Beam earned his PhD at North Carolina State University in 2014, working on Bayesian neural networks. He subsequently completed a biomedical-informatics postdoctoral fellowship at Harvard Medical School and served as a junior faculty member. His industry roles included Senior Fellow at Flagship Pioneering, leading machine learning across its portfolio, and founding head of machine learning at VL56, a Flagship-backed protein-engineering venture.[3][2]

At Harvard, Beam’s group has pursued predictive tools for neonatal intensive care using imaging, clinical notes and real-time monitoring data. Its research aims include combining those modalities and studying how models can fit naturally into clinical workflows. The group has also worked on extracting medical and diagnostic principles from unstructured text to support tasks such as differential diagnosis and answering medical questions.[3]

A central distinction in the group’s research is between predicting an outcome and choosing an intervention. Its Harvard research description puts the clinical need plainly: “what doctors often want are decisions”. In collaboration with Harvard Chan’s causal-inference colleagues, the group explores methods that can account for intervention effects. Beam also co-founded NEJM AI, with the aim of helping define how AI is evaluated and adopted in clinical practice.[3][2]