Where LLMs actually add value: a pre-registered intervention that fixed the wrong mechanism
Six months of audited live evidence from a multi-LLM investment committee, including the negative results.
Research
From pure mathematics to pre-registered AI studies.
The same habit runs through all of it: state the claim precisely, test it honestly, and publish what the evidence says, including when the answer is no.
Working papers
Evidence on where AI actually adds value.
Field notes from open systems, written so their claims can be inspected before results are added.
Grounded, not automated: a verification-first LLM career system
Separates demonstrated capability from process and outcome claims, with a prospective evaluation protocol.
Studies and briefs
Pre-registered, reproducible, and honest about negative results.
Research builds with frozen protocols, control conditions, and confirmatory results kept apart from post-hoc explanation.
Can a little graph theory make grid-control AI learn faster? A pre-registered no.
A solver-free connectivity test, grounded in the graphic matroid, can screen power-grid topology actions far more cheaply than simulating each one. I pre-registered a study to see whether that screen also helps a reinforcement-learning agent learn faster. It did not, and the reasons are worth more than a positive result would have been.
Can three models argue their way to a better investment decision?
ZenInvest let a committee of Claude, GPT, and Gemini debate every trade on a paper-trading account for six months. The honest answer was no. This brief covers what the evidence showed, why it was published, and how the redesign, ZEN-2, is built to measure LLM value rather than assume it.
Can an agent get better without changing its weights?
ZenArena puts one of agentic AI's favourite claims — that memory makes agents better — under a falsifiable test, using chess and a Stockfish truth signal to measure whether governed memory beats remembering everything.
Mathematics
Peer-reviewed research in pure mathematics.
PhD research at the University of Exeter on Hopf-Galois theory and skew braces: algebraic structures that generalise classical Galois theory and connect to solutions of the Yang-Baxter equation.
Skew braces and Hopf-Galois structures of Heisenberg type
Classifies all skew braces of Heisenberg type for primes p > 3 and the corresponding Hopf-Galois structures.
On Hopf-Galois structures and skew braces of order p³
Classifies all Hopf-Galois structures and skew braces of order p³.
Some braces of cardinality p⁴ and related Hopf-Galois extensions
with D. Puljić and A. Smoktunowicz
Talks and recognition
Sharing the work beyond the code.
Speaking, advisory, and recognition that connect research practice to enterprise AI.
Independence
A note on this research.
The working papers, studies, and briefs here are personal, non-commercial research maintained in a personal capacity, independent of my professional employment and not affiliated with any employer. The mathematics publications are academic work from my doctoral and postdoctoral research.