Building research infrastructure that scientists can actually use

Much of my work is motivated by a practical question: how do we turn powerful but fragile research workflows into tools that other scientists can trust and use?

That question shows up in several forms. Sometimes it becomes an exascale software problem. Sometimes it becomes a workflow-automation problem in chemistry. More recently, it appears as an AI-for-science question around agents, retrieval, evaluation, and tool use.

The common theme is scientific leverage. Better infrastructure does not replace science. It changes what kinds of science become routine.