Bowie Duncan

Owned end-to-end data science for production ML and risk scoring, including continuous training, deployment, monitoring, and failure analysis at very large scale. Led causal inference and holdout FDR experimentation with stratified sampling and live measurement to estimate intervention effects and guide decisions. Built and evaluated GenAI/agent workflows with guardrails, human-in-the-loop calibration, and error analysis, and operated RCA and KPI measurement to improve risk detection and decision quality.

Conducted quantitative research using causal inference and DAG-based experimental and observational study design. Performed statistical analysis and reproducible workflows using R and tools such as SPSS and JASP, supporting multiple multi-institution and NIH-funded studies. Co-authored peer-reviewed publications and helped communicate results for multidisciplinary collaborators through analysis, validation, and publication cycles.