At Path2Response LLC, I developed and retrained client-specific XGBoost response-propensity models across 20+ accounts. I also built Python and SQL workflows to prepare campaign, response, and browsing data for predictive modeling.
To assess ranking stability across heterogeneous client models, I independently designed bootstrap experiments for a collaboratively developed MCMC training-data resampling methodology. The experiments showed average Kendall’s τ rising from approximately 0.60 to 0.80 relative to the original sampling approach.
At the University of South Carolina, I developed Bayesian hierarchical models for diagnostic accuracy from multi-reader ordinal ratings and implemented Gibbs samplers for posterior inference. I’m now developing a five-stage LangGraph analytics agent with a statistical validation layer and an evaluation pipeline.
