Modeling Engine Interfaces & Infrastructure
RsNLME (R Speaks NLME)
CRAN-distributed R interface to Certara’s NLME engine. Enables complete pharmacometric workflows in R — including model specification, parameter estimation, simulation, and diagnostics — across FOCE, Laplacian, and AGQ algorithms.
Designed as production-grade infrastructure bridging statistical computing (R) with compiled estimation engines.
Key Contributions
- Engine wrapper architecture
- Cross-platform execution orchestration
- PML-based model specification pipeline
- Shiny-integrated interactive workflows
- CRAN-compliant package engineering
Technologies: R, NLME engine, PML, S4/R6 OOP, Shiny
Optimization & Automated Model Search
pyDarwin (Machine Learning–Driven Model Selection)
Architectural contributor to pyDarwin, a machine learning framework for automated population PK model selection integrating NONMEM and NLME engines.
Designed and validated large-scale search infrastructure spanning 1.57M model combinations to ensure reproducible global optimum discovery.
Supports multiple optimization strategies:
- Genetic Algorithms (GA)
- Gaussian Processes (GP)
- Random Forests (RF)
- Gradient Boosted Regression Trees (GBRT)
- Particle Swarm Optimization (PSO)
This work formalizes algorithmic model search as reproducible scientific infrastructure rather than heuristic exploration.
Technologies: Python, R, NONMEM, NLME, machine learning algorithms
Simulation Diagnostics & Visualization Systems
tidyvpc
CRAN-distributed R package implementing simulation-based Visual Predictive Checks (VPCs) for NLME models.
Provides a structured, tidy interface for generating prediction-corrected, stratified, and binless VPCs with reproducible simulation diagnostics.
Technologies: R, tidyverse, simulation workflows
vachette
CRAN-distributed R package introducing a novel covariate-harmonized visualization methodology for pharmacometric models.
Implements variability-aligned, time-transformation equivalent (vachette) plots that unify observations into a single reference-aligned diagnostic framework.
Published in The AAPS Journal (2025).
Technologies: R, ggplot2, NLME diagnostics
CRAN
DOI: 10.1208/s12248-025-01131-9
Integrated Modeling Workbench Systems
Pirana Shiny Application Suite
Architect and contributor to interactive modeling applications integrated into the Pirana pharmacometric workbench.
Applications include:
- Model Builder — graphical NLME model construction and PML generation
- Model Results — diagnostic plots, tables, GOF analyses, and report shell generation
- VPC Results — interactive visual predictive check interface
Designed modular Shiny components interfacing directly with modeling engines and automated workflows.
Technologies: R Shiny, RsNLME, ggplot2, Pirana APIs
R Ecosystem Infrastructure
Certara.R
Central hub for open-source pharmacometric R packages developed within the Certara ecosystem.
Provides:
- Structured package distribution (including non-standard repositories)
- Documentation via pkgdown
- Interactive tutorials via learnr
- CI-driven build and release pipelines
All major packages (RsNLME, tidyvpc, vachette, XposeNLME) are publicly distributed and CRAN-compliant where applicable.
Technologies: R, pkgdown, learnr, GitHub Actions, CRAN infrastructure
Cloud-Native Modeling Platforms
Contributions to next-generation cloud-native pharmacometric modeling platforms within the Phoenix ecosystem.
Systems include model configuration environments, simulation orchestration layers, and modeling workflow management delivered via modern full-stack architecture.
Key Areas
- React/TypeScript front-end systems
- Node/Express and Java-based backend services
- API integration with modeling engines
- Containerized execution environments
- OAuth/OIDC-aware identity integration
- CI/CD-enabled deployment pipelines
These platforms translate complex modeling infrastructure into scalable, production-grade cloud systems.
Technologies: React, TypeScript, Node/Express, Java, Docker, OAuth/OIDC