Welcome!
I build AI systems. Not slideware about AI, and not forecasts of where it is headed. Working systems that run in production and handle real, confidential data every day.
Over the past decade that has meant training local models to read 1.3 million child welfare case narratives, building a de-identification model that outperforms the standard commercial tools, standing up secure document-intelligence systems for legal matters where nothing can touch the cloud, and designing decision-support dashboards now in daily statewide use.
I benchmark models against published, peer-reviewed evidence, and I have delivered training on this work to professional audiences nationally and internationally.
Who am I?
I am a Professor at the University of Michigan School of Social Work and a co-founder of the Child & Adolescent Data Lab. I run Parallel 42 (P42), a fractional AI advisory practice for organizations with high-stakes data in human services and law. I teach in the AI Law and Policy Clinic at the University of Michigan Law School, and I keep an active expert witness practice supporting large-scale document review. Twenty years of applied research stand behind this, ten of them in AI.
What you’ll find here
This is where I write about the work for the people who depend on these systems without building them, in both social work and law. Some posts are field notes from something I just built or tested. Others make an argument about where the field should be heading, or explain a technical idea in plain language for a professional who needs it. When something is uncertain or contested, I say so. The through-line is the case for a different default: small, local, and accountable AI, built by and answerable to the people who use it. That is what this newsletter is for. Building the alternative, in the open.
