Building with AI,as infrastructure.

In active design

AI Learning Companion

A platform concept for a persistent AI agent that supports a student across their full journey, from first inquiry through graduation and beyond. It grew out of a clear gap: instructors had little insight into how students were actually engaging, and students had no continuous point of support between touchpoints. Designed to remember context over time and adapt as a learner moves through stages, beginning with advisory support before taking on more active capabilities.

Shipped

Custom AI Tools for Education

Two focused tools built for Keyin's programs. One helps prospective students explore program options and figure out what fits them; the other helps enrolled students catch up on material they've missed. Both use pedagogy-aware prompting tied to different learning preferences, rather than generic chatbot behavior. Beyond their direct use, they surfaced the engagement-visibility problem that motivated the learning companion platform.

Infrastructure

AI-Accessible Data Layer

Infrastructure that lets AI tools answer questions grounded in real institutional data instead of guesswork. The goal was a governed interface where an agent can retrieve accurate, current information about programs and the people moving through them, with the guardrails to keep that access safe and scoped. This turns institutional data from something locked inside dashboards into something conversational systems can actually reason over.

Shipped

Internal AI Development Platform

A system for spinning up and deploying internal applications quickly, built so a small team can ship production tools at the pace the AI era demands. It standardizes the repetitive parts of getting an app live so the team spends its time on the actual product, moving experiments from idea to running service in days, not quarters.

Daily practice

AI-Native Operating Model

The practice underneath everything else. I work as an active engineer alongside my product role, using AI coding agents as a core part of how I build: parallel development workflows, codified instructions that give agents the right context, and a repeatable personal system for getting real leverage out of these tools rather than novelty. I treat my own workflow as a system to optimize; a lot of what I design for the institution starts as something I've tested on myself first.

Advisory

AI Adoption Advisory

Advisory work with organizations adopting AI tooling across engineering and operations. The recurring lesson is that the hard part is rarely the tool itself. It's change management, workflow fit, and culture. I help teams figure out where AI genuinely accelerates their work and where it adds friction, then sequence adoption around that instead of forcing it everywhere at once.