Software
Released packages, PostgreSQL-native by design. Each keeps the computation inside the database that already holds the data, so a pipeline reaches production without moving anything out of it.
Temporal machine learning for public policy, end to end inside PostgreSQL. A fork of the University of Chicago’s Triage in which cohorts, labels, features, temporal cross-validation and model governance are all expressed as SQL against the database that already holds the data, so a pipeline moves to production without leaving it.
Deep Feature Synthesis compiled to SQL. Given a relational schema and an entity graph, it generates the point-in-time-correct feature matrix as queries the database runs itself, so the database computes each feature as of a date rather than joining it after the fact.