Energy Economics • AI Infrastructure • Applied Econometrics
I'm an applied economist focused on energy markets and AI infrastructure. I'm completing an M.S. in Applied Economics at the University of Maryland and hold an MPA from the University of Montana. My research sits at the intersection of electricity systems, data center investment, and the limits of what current grid planning models can anticipate.
Before moving into economics I spent more than a decade running operations in hospitality. That background shaped how I think about uncertainty and decision-making under real constraints — where the model is only as useful as the decision it supports, and where being wrong has immediate consequences.
My primary research asks how AI infrastructure investment is reshaping regional electricity demand — and whether grid planning tools are built to see it coming. The empirical paper, AI Infrastructure and Regional Electricity Demand: Evidence from U.S. Interconnection Queues, estimates a 34.8 index-point divergence in ERCOT minimum hourly demand above a synthetic counterfactual using a four-layer identification strategy across PJM, ERCOT, and MISO. It also documents the absence of public load-side queue data as a policy finding.
The companion policy brief, The Missing Half of the Queue, makes the regulatory case directly to FERC: require grid operators to publish large load interconnection queue data on the same terms currently applied to generation. The brief traces thirty years of FERC precedent and proposes a specific mechanism that does not require new legislation.
Outside the research, I build software. Bearing is a mobile-first daily intelligence tool I designed and built solo — React, Supabase, the Claude API, Stripe, and Vercel, from zero to a deployed product with auth, payments, and cross-device sync. Building production AI applications clarified how these systems behave at the application layer in ways that inform how I think about the infrastructure side of the research.
I build end-to-end — data construction, model design, identification strategy, diagnostics, and interpretation. The code is part of the deliverable. Every analysis is documented so the reasoning is visible, and every significant result comes with an explicit account of why clean identification is or isn't possible with available data.
I use Python, R, SQL, and Stata, and I'm comfortable moving between econometric work and production software depending on what the problem requires.
You can reach me through the contact form or on LinkedIn.