Tools that ship, not just papers.
From data science to machine learning to AI: the frontier keeps moving, and the tools follow the public there. That has meant building an array of analysis engines, including for media monitoring, adverse event discovery, and health marketplace surveillance. Most recently, I was one of the principal architects of an automated AI scientist platform.
Tobacco Watcher
Automated media surveillance for global tobacco control: 350,000+ sources monitored around the clock, operating natively in 23 languages, every language spoken by at least 1% of the world. Actionable intelligence, pushed to advocates and decision makers.
ChatCPR
An open-source AI CPR instructor that outperformed 911 dispatchers and every frontier model tested, including ChatGPT, Gemini, Llama, and Grok. Built on open models so it can be deployed anywhere, without per-use costs.
Waldo
An open-source AI that reads unstructured self-reports and finds the adverse events official safety databases never see. Waldo caught adverse events with 95% accuracy and produced 18-fold fewer false alarms than ChatGPT, then scanned 437,000+ posts to surface nearly 29,000 potential adverse events from unregulated cannabis products alone.
Choice Watch
An analysis engine that archives and indexes crisis pregnancy center websites nationwide. Choice Watch discovered 1,825 centers operating across the US and showed 91% advertise medical services, including scientifically unsupported abortion pill reversal on nearly a third of sites. It turns anecdote into evidence for clinicians and rulemakers.