Systems I've shipped
Products, research surfaces and open-source tooling — each one linked to the live thing, not a rendered screenshot of it.
Microx
Research synthesis workspace · Android + web
Multi-model research workspace: parallel agents read sources, surface consensus and disagreement, and produce a cited expert report. Shipped on Android with a web companion.
Mneme
Private, memory-first AI twin · Android
On-device-first personal AI that keeps a long-lived memory of your life, health signals and conversations, and answers from that memory instead of from a generic model.
Coordination visualiser
Agent-systems lab
A visualiser for how agent populations coordinate: how work splits, where handoffs stall, and what a swarm actually does when the task graph gets deep. Built to make agent behaviour legible rather than magical.
23 public repositories
Tools, notebooks and experiments
From agent tooling and evaluation harnesses to data and ML notebooks — everything I build in the open lands here, with the reasoning in the README rather than a marketing page.
Five publications
Peer-reviewed research
Decision science, optimization, biomedical AI and molecular dynamics — including the enzalutamide drug-resistance work in the Biophysical Journal. Every entry carries its DOI and citation count.
How engagements start
I take a small number of engagements each quarter — tell me the problem and the timeline and I'll tell you honestly whether I'm the right person for it.
Scoping conversation
A short call or an email through the contact form: what the system has to do, what exists today, and what a good outcome looks like.
Written proposal
You get scope, deliverables, timeline and commercial terms in writing before anything starts — including how the work is invoiced.
Build in the open
Work runs in short increments with evaluation gates and a written handover, so your team can keep operating the system after I step back.