Bring the Data to the User
An NHANES MCP server that gets the survey weights right, validated against published NCHS estimates, and what the pattern means for registries, EHR and claims data.
On causal reasoning, AI in research,
and the craft of building better evidence.

A proof of concept: an open-source MCP server that lets an agentic AI assistant analyze NHANES the way NCHS does, checked against published estimates.
An NHANES MCP server that gets the survey weights right, validated against published NCHS estimates, and what the pattern means for registries, EHR and claims data.
Why domain expertise—and a willingness to experiment—can be the starting point for useful AI in pharmacoepidemiology.
Four prompting strategies, two model families, one research task. What changes when models interpret more of your intent?
What a kitchen brigade can teach us about dividing work, preserving context, and building AI systems for research.
A practical way to ask whether a study’s design can answer the causal question its authors are asking.