A research programme built on one question

How can we design intelligent systems that adapt, optimize and create, not just compute?

My answer is generative optimization: AI that generates, models of engagement that evaluate, and metaheuristics that search, built up since 2012.

Behind this page is a live latent space: every point is an embedding, not a word, and each spark is a thought moving through it, attending to its neighbours as it goes. Click anywhere to start one.

/ books and tools

Things I have built that you can use.

A free book to start with today, and a writing app and a second book on the way.

/ the loop, running live

Watch an algorithm design, on two different problems.

A generator edits Sudoku puzzles, a simulated player rates how engaging each one is, and a search archive keeps the best puzzle for every style. Most designs land too easy; watch the search find the narrow comfortable zone.

GENERATE→EVALUATE→SEARCH↺
CANDIDATE
GAME REFINEMENT · GR = √G / T
too hardcomfortable 0.07–0.08too easy
waiting for the first candidate…
MAP-ELITES ARCHIVE · click a niche
symmetry →
22 givens← fewer clues · more clues →45
comfortabletoo easytoo hard
0
Puzzles evaluated
0 / 40
Niches filled
0
In comfort zone

An illustrative teaching model of my work on puzzle generation and game refinement: the generator here makes random edits where the research uses AI agents, and GR = √G / T is a simplified measure (G cells placed, T candidate values weighed).

/ research programme

Generative optimization: one loop, three stages.

Simulation is the test bench. 94 of 113 publications use agent-based, self-play or Monte Carlo simulation to run the loop before anything reaches real players, patients or factory floors.

NOW
Generative Optimization
Closing the loop for individual design problems, optimized against human engagement.
NEXT
Generative Optinformatics
A loop that learns from its own search history and transfers it to new design problems.
Read the programme →

/ selected work

Selected articles.

All 113 publications →

Open to collaboration.

PhD and Master's supervision at every stage of the loop, industry research collaboration, and conference invitations.

Get in touch →