Software & Code

A selection of research-grade tooling I’ve built. Repositories are currently private; available on request.

temporal_wgan

A PyTorch package for treatment-effect estimation in panel data via temporal Wasserstein GANs. Sig-WGAN, TimeGAN, and TimeGrad backbones; difference-in-differences and CATE estimators; identification-theory utilities. Companion code for “Conditional Average Treatment Effects as Dynamical Systems.”

Hansard NLP pipeline

End-to-end pipeline (segment → embed → classify → assemble) for 13.1M+ sentence segments of UK parliamentary speech (2000–2025), with comparable US Congressional corpora. 7-class emotion classification over nli-mpnet embeddings with a fine-tuned semantic adapter. Companion code for “Moral Sentiments in Parliamentary Discourse.”

Causal Geometry validation suite

Synthetic experimental framework for the Causal Geometry proposal. Intrinsic-dimension estimation (Two-NN), tensor decomposition (PARAFAC), and Gaussian-mixture clustering over an 11-dimensional geometric feature space; validated on Lorenz, pendulum, and coupled-oscillator systems.

Mexican professionals data infrastructure

Production-grade scraper and enrichment pipeline (Wikidata, SEP Solr) for the universe of ~1.2M Mexican credentialed professionals. Includes Bayesian missing-data handling, a proxy/worker queue system, and a Dockerized deployment. Companion infrastructure for “Political and Economic Spillover Effects of Brain Drain in Mexico.”