JD Pruett

I lead engineering at Varick, building the product and ML systems behind enterprise agent deployments.

The work is less sexy than Twitter might have you believe. Building good ML systems in the enterprise is largely an engineering challenge, finding the right blend of durable execution, data science, permissions, and edge-case handling.

Nonprofits compete for the same funding, but that competition doesn't reliably produce differentiation or efficiency. Collaboration would help, but the cost of finding partners and coordinating work generally outweighs the benefit. Euterria seeks to lower the collaboration cost.

Built for Bay Area climate nonprofits, Euterria indexes knowledge scattered across the network, from who's best at what to practices buried in grant reports and failed experiments in internal decks. It turns that knowledge into a network-wide internal search engine and collaboration platform. We're funded by Stanford's TomKat Center and piloting with around thirty organizations around the Bay.

The Design & Technology Institute is a vocational school in Accra founded by the ever-inspirational Constance Swaniker. DTI and its industry partners train thousands of young Ghanaians in skilled trades and are doing the hard work of scaling up their programming nationwide.

As a SEED fellow, I had the privilege of working with DTI to ensure they had the data management and technical systems in place to speed up its incredible work around the country. That meant a combination of finding the right off-the-shelf software and deploying it around the country, and building custom tools for things like procurement, inventory, and student surveys.

The Political Psychology Research Group is a cross-disciplinary lab bringing together economics, psychology, and political science to study the psychological underpinnings of political behavior. Jon's group publishes heavily on the deceptively difficult practice of running and analyzing reliable surveys for social science.

I led a small team to do a multi-year independent evaluation of sample matching, an emerging method of nonprobabilistic (non-random) survey sampling beginning to consume the industry. Across 17 years of data, we systematically examined one of the most cited datasets in political science, testing for changes in accuracy over time and deconstructing the error to reveal how good this kind of survey really is.

Russ' lab studies the neural basis of adaptive behavior, and is a leader in open neuroscience. My corner of it was cognitive control: inhibitory control and cognitive flexibility, the machinery you use to stop yourself and to change course. It's the part of the mind that fails us in many of the behaviors we most want to change.

I worked on a multimodal imaging study asking whether we could find the neural and behavioral markers that predict how people with putative deficits in cognitive control respond to an intervention meant to strengthen it. We ran it as a pre/post trial in people managing binge eating and nicotine dependence, pairing task-based fMRI with a mobile intervention. I designed our mediation analysis strategy to causally estimate the neural signals that actually carried the intervention's effect on behavior.

Steve Ballmer is the former CEO of Microsoft, and his foundation is a large philanthropic funder in the LA, Detroit, and Seattle metro areas. After I presented to Steve in a class, I worked with him and his philanthropy team to figure out what it would take to offer afterschool programming to every kid who wanted it across the three metros.

I led a three-city study of the afterschool landscape in collaboration with local government and NGOs, and developed a cost model for afterschool education. The work culminated in a $450M annual budget that leaned on existing public infrastructure and a number of investments in youth extracurricular education.