About
I am a machine learning engineering leader with 21 years of experience building production ML and data systems, including 12 years leading engineering teams. My work has focused on building and operating systems where data correctness, model evaluation and production reliability matter.
This site documents independent research and engineering work outside my professional role, across three projects: Finch, an ML research platform built to be causally correct and reproducible; Crow, a language and runtime designed so that correctness is enforced rather than assumed; and BlueJay, an earlier experiment in data storage and indexing. What connects them is an interest in systems that make incorrect results detectable, and the writing reports what was built and what did not survive scrutiny.
On authorship: Finch and Crow were implemented largely by AI coding agents. I owned the architecture, the research questions, the correctness criteria and the decisions about what to build, accept, reject or stop.
