The numbers
The portfolio in numbers, actually computed from the site's own data on every build, not typed by hand into a template.
Downloads per month on PyPI
Sum across every command-line tool published on PyPI, measured with pypistats.
Open-source tools
Command-line utilities for the daily work of ML, published and maintained under jmweb-org.
End-to-end projects
From raw data to a served model: credit risk, churn, RAG, a from-scratch transformer and more.
Test functions
Unit and integration tests with pytest, counted by hand across the tools and the projects.
Blog articles
Technical notes written while learning, no filler.
Interactive demos
The lab experiments plus the tokenmeter, splitcheck and pii-sweep demos, all running in the browser.
Site languages
Spanish and English, with hreflang and content translated by hand on every page.
Trained checkpoints
From the transformer built from scratch: the base model, the BPE tokenizer, and two ablations, char_1head and char_nopos.
Real numbers as of the site's latest publish; downloads come from pypistats and refresh on every deploy.