Research
My research runs along two connected strands. The first is bias-robust Bayesian evidence synthesis: reproducible statistical workflows for applied literatures where publication bias, small-study effects, and selective reporting make conventional summaries difficult to interpret. The second is human-governed AI infrastructure for curriculum-scale knowledge work: systems that keep large, AI-assisted course production coherent, bounded, reproducible, and reviewable, with instructor judgment in control.
Bayesian evidence synthesis
Quantifying Evidential Rigor in Meta-Analytic Corpora
A Simulation-Characterized, Bias-Robust Bayesian Workflow with a Nutrition Case Study.
This preprint introduces a corpus-scale Bayesian evidential-audit workflow for meta-analytic evidence. It fits a matched Bayesian random-effects baseline alongside a bias-aware, model-averaged ensemble, then summarizes evidential yield with a rigor estimand: resolved evidence for an effect or for no effect, together with the absence of an explicit modeled bias component. The workflow is characterized with simulation and resampling designs and demonstrated on a nutrition intervention meta-analysis corpus. The companion repository supports reproducibility and adaptation of the workflow to other literatures.
Course Builder
Course Builder is a developing course-design harness — a repo-native analytic exoskeleton for building course materials. Rather than improvising from an open-ended prompt, a run extends a course as an explicit analytic object under a set of controls: the course context, source boundaries, pedagogy references, durable state, diagnostic gates, and human review.
It is not an autonomous “make me a course” generator. Instructor judgment and release authority remain central: the harness prepares and proposes work and records honest status, but a person verifies correctness, confirms source licenses, signs off on accessibility, and authorizes any release.
The ten-site teaching portfolio is Course Builder’s current public demonstration layer. The public course sites carry durable notes, labs, worked examples, and curated resources; roster-specific, graded, answer-key, and instructor-only materials stay in the LMS or in private workspaces, never on the public sites.
Course Builder is still under active development. It has not been empirically validated, makes no claim about learning gains, and its working manuscript is a developing draft rather than a publication.
Research interests
Bias-robust Bayesian meta-analysis. Model-averaged Bayes factors, selection models, heterogeneity, and evidence for or against effects after accounting for bias.
Publication bias and evidential fragility. Selective reporting, small-study effects, threshold behavior, and how these distort applied literatures.
Corpus-scale evidential auditing. Summarizing evidential yield across many meta-analyses, rather than one effect at a time.
Nutrition and health evidence synthesis. Applying bias-aware Bayesian methods to intervention meta-analyses in nutrition, fitness, and health.
Reproducible statistical workflows. R, Quarto, Git, simulation studies, public repositories, and transparent research pipelines.
Human-governed AI and curriculum infrastructure. Context-governed systems for AI-assisted course production that keep generation traceable, bounded, reproducible, and under instructor control.
Statistics education and reproducible course systems. Introductory statistics, Bayesian intuition, durable public course materials, and the responsible, transparent use of AI assistants in mathematical and statistical learning.