projects

Research software for scientific computing, Bayesian inference, and operational deployment infrastructure.

lead developer · released
VBPCApy workflow from an incomplete matrix and explicit observation mask through variational fitting, diagnostics, component selection, and predictive uncertainty.

vbpca-py

Variational Bayesian PCA for incomplete data with full posterior uncertainty, calibrated predictive variance, automatic component pruning, native missingness handling, and C++-accelerated kernels.

trade-study workflow from a constrained design space through simulator protocols and scoring to Pareto and surrogate decision support.

trade-study

Design and evaluation framework for scientific simulation studies. Score competing configurations — model formulations, solver choices, measurement strategies, or any design decision — against known ground truth via protocol-driven simulators, proper scoring rules, hierarchical phases, multi-objective Pareto optimization, and Bayesian model stacking.

lead developer · in development
Predictive accuracy and structural admissibility shown as complementary evaluation criteria for scientific forecasts.

Current Model-Analysis Portfolio

Active mechanistic modeling, inference, and evaluation studies spanning dengue, COVID-19 observation noise, diphtheria vaccination, CCHF surveillance, and structural scores.
OP Engine workflow connecting prepared models, numerical kernel families, backend-native providers, and execution diagnostics.

OP Engine

Array-API-polymorphic ODE/PDE solver with portable numerical kernels and backend-native execution providers for explicit, IMEX, implicit, and stochastic integration.

OP System compiler workflow from equations or transition diagrams through validation, typed intermediate representation, and portable evaluators.

OP System

Restricted expression language and compiler for structured dynamical systems, lowering model specifications through typed IR to vectorized, Array-API-polymorphic code.

pp-eigentest fit-check-select workflow using masked model fitting, predictive diagnostics, and posterior-predictive spectral rank selection.

pp-eigentest

Private pre-release work on posterior-predictive signal-rank selection, centered on a sequential fitted-model selector with explicit calibration and stopping diagnostics.

collaborative infrastructure
FlepiMoP2 workflow from validated configuration and plugin resolution through campaign execution and persisted outputs.

FlepiMoP2

Configuration-driven orchestration engine for CDC-supported infectious disease forecasting and scenario analysis. Plugin architecture decoupling model specification, integration, and persistence.