Lead with vision, build with teams.
Direction first, then the people, architecture, and rituals that turn it into delivery. The strongest AI work is a team sport.
Mathematics · agentic systems · public R&D
Building AI that makes humans more capable, curious, and in control.
I build and write about agentic AI systems that combine model reasoning with deterministic controls, evaluation, and human judgment. This site is the public record of that work.
- Open-source systems
- Evidence-led R&D
- Human control at the boundary
Public work, built to inspect
Open systems, testable ideas, and honest caveats.
The open-source systems and research briefs publish code, architecture, safety posture, and lessons instead of hiding them behind product claims.
How the work happens
Three habits carry ideas into working systems.
Mathematical rigor, capable teams, and clear teaching keep ambitious AI work useful and accountable.
Apply mathematical rigor to AI and data.
Models that earn their conclusions and systems that stay legible as they grow. Data science, agentic design, and evaluation grounded in clear reasoning.
Teach what we learn.
Writing, lecturing, and public work that turn experience into something others can use. Education and transformation are the same skill in different rooms.
Featured project
ZenInvest is the flagship open project on the site.
It brings together agentic design, deterministic controls, and human oversight in one of the hardest consumer AI domains: investing.
featured
ZenInvest
An open, agentic investing system with deterministic controls and clear operator oversight — currently a proof-of-concept running on Trading 212's Practice (paper) account, so no real capital is at risk.
Why it stands out
- Agentic architecture that stays inspectable
- Operator-supervised autonomy
- Learns only behind hard gates
- Technical product posture over return promises
From the lab
R&D briefs: agentic AI put under real constraints.
Short write-ups of the systems built in the lab — each one a testable idea, the engineering behind it, and the pattern worth reusing.
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.
Agents advise, actuaries decide: an AI pricing engine a regulator could read
ZenRate wraps a deterministic actuarial core in a layer of collaborating AI agents — where exactly one agent holds a veto, every decision is recorded bi-temporally, and the compliance paperwork falls out of the architecture.
Forecasting as a loop, not a number on a slide
ZenForecast is an open-source, governance-first Python framework that treats forecasting as a continuous predict → decide → act → learn loop — with interchangeable models, leakage-free backtesting, and an audit trail behind a single API.