My PhD is about one question: how do you trust the output of an autonomous agent? The answer I work on is a proposer, challenger, judge architecture, where independent agents argue and adjudicate before a decision is acted on. The same chassis I study academically is the safety layer I deploy in real systems (OC1, OC2, MVF-Composer).
The single-LLM-call pattern is convenient and dangerous: convenience because it is one prompt, dangerous
because the verifier is the same model that produced the claim. My dissertation develops a multi-agent
pattern where a Proposer drafts a decision, a Challenger attacks it, and a Judge commits a verdict. The
transcript is hashed and committed externally, so the audit trail is outside the LLM rather than inside
it.
The pattern shows up in three deployed systems (OC1, OC2, MVF-Composer) and in the IEEE ICBC 2026 paper
on stablecoin reserve control. I treat the engineering and the research as the same feedback loop: I
ship the system, instrument it, find the failure mode the paper didn't predict, and ship the next paper.
Themes: multi-agent systems, consensus and trust, calibrated forecasting, blockchain and
decentralized-system safety.
Patterns where independent agents argue, judge, and commit a decision outside the LLM that drafted it.
Decisions the system can defend with a measurable confidence score and a human-readable audit trail.
Defense mechanisms for the moments when a stablecoin depegs, replayed block by block against real crises.
Why trust assumptions break, when cryptography helps, and when you still need a human in the loop.
Four publications, single-author first author on the most recent, an invited talk at IEEE ICBC 2026.
University of Notre Dame · 2020 to 2026 (expected)
Advisor: Jarek Nabrzyski
Purdue University · 2018 to 2019
University of Washington · 2014 to 2018
Happy to share preprints, code, and replication notes under standard academic terms. I read every message.