<> AI Research @ UVA · Faculty AI Upskilling Workshop · Delivered July 20, 2026
Designing an AI-Powered Research Pipeline
You didn’t learn Excel or R by collaborating with them — you dictated every step. Generative AI rewards the opposite instinct, and that’s the under-recognized sticking point: you don’t learn to use AI, you learn to interact with it.
> A two-hour, in-person workshop for faculty, PhD students & research staff in any discipline — no coding or prior AI experience assumed. This page is a record of the session as delivered.
AI Research @ UVA asked for a session that would help established researchers — in any discipline, at any level of technical comfort — actually get somewhere with generative AI. This page records what was asked for, what the faculty who enrolled said they needed, how the two hours were designed around that, and what was shown. The full deck is at the bottom.
The Ask
AI Research @ UVA — the Provost’s office initiative co-directed by Henry Kautz and Ryan Wright — runs a workshop series to build AI capability across the University. The invitation was to design and teach a session in that series: two hours, in person, open to faculty, PhD students, and research staff from any school.
The brief made it a design problem rather than a training problem. A single session had to work for a room that would arrive with wildly different technical backgrounds, different disciplinary methods, and different institutional constraints on what they could even put in front of a model — and it had to leave each person with something they could use on Tuesday. That is the same problem the design-sprint method is built to handle, which is why the session was designed as one, rather than as a tools tutorial.
What Faculty Said They Needed
Enrolled participants were surveyed before the session. Eight of the twenty responded — enough to steer the design, not enough to be representative, and treated that way. Three findings shaped every decision that followed.
The clearest shared need was not a deeper technical course. It was a repeatable way to move from a research problem to a productive AI experiment — and a different way of working with the model once you got there.
Two further signals mattered as much as the headline numbers:
- The room was mixed, not uniformly non-technical. Every respondent had worked successfully with conversational prompting, but only a minority had used AI for coding or data analysis, structured extraction, agents, or building prototypes. A beginner’s tour would have wasted the experienced half; a technical session would have lost the other.
- Most respondents carried a hard constraint. Unpublished or sensitive findings, HIPAA, IRB limits, sponsor and IP restrictions, FERPA — a majority named at least one. Whatever was demonstrated live had to be something they could legitimately practice on.
Source: pre-workshop survey of enrolled participants, reported as de-identified aggregate findings.
How the Session Was Designed
Six deliberate moves, each answering something the survey surfaced.
The Content
The spine of the session is a five-stop pipeline. Input: your domain expertise. Output: a pipeline you own. The design loop — diverge, converge, repeat — runs at every stop, and verification runs underneath all of them.
AI doesn’t replace your expertise — it makes it buildable.
The Two Hours
Who Was in the Room
Established researchers with deep domain expertise, from across the University. The research terrain represented in the pre-workshop responses alone spanned patient decision support, how AI changes human decision-making, consumer behavior and fairness, administrative workflow synthesis, survey design and truthful disclosure, large-scale environmental and health data, undergraduate social networks, and mathematics research supported by code and pattern search.
Built to Send People Home With
- AI-era research questions from their own domain — including ones previously ruled out as infeasible
- A starter context library — expertise and question structured so an AI system can work with it
- Interactive prototypes exploring where those research trajectories might lead
- A personal roadmap for an AI-powered research pipeline — and a clear sense of where traditional methods re-enter
How This Is Grounded
The workshop was designed as a human on-ramp to the goals of AI Research @ UVA — to “cultivate the next generation of AI researchers and practitioners” and move the University from “isolated pockets of excellence to a unified Networked University.”
The method behind it is a design sprint: define the right question, ground the AI in your expertise, prototype fast. It’s the same approach that powered the Health Design Sprint, where eight medical students with no coding background shipped working clinical tools in ten days — built on the UVA Medical Design Program, which has trained 200+ students in human-centered design since 2017.
Health is where the method was proven, but the way of working — and thinking — transfers to any discipline: a historian structuring an archive, a nurse redesigning a care workflow, an economist exploring a new dataset.
An Associate Professor of Emergency Medicine and Public Health Sciences at the UVA School of Medicine, and Director of the UVA Medical Design Program — which has taught human-centered design to 200+ clinicians since 2017.
His work focuses on putting design thinking — and now generative AI — directly in the hands of domain experts, so the people closest to a problem can build for it.
The full deck, with annotated slide-by-slide notes.
A podcast episode with Matt Trowbridge released alongside the workshop.