The Right Trial, for the Right Patients
At the RECOVER-TLC workshop, we argued that health records and AI can test a treatment idea in weeks, before a trial enrolls its first patient
Research on agentic clinical AI, temporal phenotyping, diagnostic data fidelity, and environmental health, from electronic health records spanning dozens of hospitals.
Each piece carries a signal strength: Signalan early idea Patterna result taking shape Findingpeer-reviewed
At the RECOVER-TLC workshop, we argued that health records and AI can test a treatment idea in weeks, before a trial enrolls its first patient
Diagnostic codes capture fewer than half of Long COVID cases across 58 US hospitals
Pythia, the multi-agent prompt optimization engine behind our clinical symptom detection work, is now open source
A multi-agent system writes, tests, and revises its own prompts to detect 72 signs and symptoms in clinical notes
ClaiD maps which conditions tend to follow which, and over what interval, across 300,000 patient records
A NAM Perspective on what health systems give up when AI spending outruns the evidence
When AI Systems That Improve Themselves Make Themselves Worse
Coordinating federal broadband funding with clinical AI could extend care to the regions with the fewest physicians
Five AI Agents That Critique Each Other — And Outperform Human Prompt Engineering