I Learn by Building
I learn best by building.
Hackathons give me a deadline, a constraint, and a reason to turn an abstract idea into something testable. The outputs may look like separate projects — Aftershock, IncidentGym, Arena — but they are all probes into the same question: how do we build trustworthy AI-native systems for resilience, security, and decision-making?
The projects
Aftershock explores how societies respond to disruption. It is a disaster-response simulation where a society of small AI agents must coordinate under pressure — triaging casualties, routing resources, and maintaining coherence when infrastructure fails. The headline finding: six small models out-deliver one big one at roughly 65% better lives-per-dollar, and written doctrine lifts protocol conformance at statistically credible levels.
IncidentGym looks at how organisations rehearse cyber incidents and measure readiness. It is a training environment where incident responders practice under realistic conditions, with deterministic scoring that makes every run auditable and replayable. The goal is not to gamify incident response — it is to make readiness measurable.
Arena asks what happens when agentic organisations and economies begin to make decisions at scale. It is a laboratory for simulating multi-agent economies where agents negotiate, trade, form alliances, and fail — providing a testbed for governance mechanisms before they touch real systems.
The common thread
The common thread is not the hackathon. The common thread is simulation, scoring, replayability, and evidence.
Every project shares a core architecture:
- Deterministic environments where scenarios can be replayed and compared
- Scoring functions that make outcomes measurable, not just observable
- Multi-agent coordination under constraints that mirror real organisational pressure
- Evidence-first reporting that distinguishes what worked from what merely looked good
These are small laboratories by design: constrained enough to build quickly, rigorous enough to test whether the underlying idea survives contact with reality.
Why public
Building in public is not about marketing. It is about accountability.
When I publish a build log, I commit to showing the negative results alongside the wins. Aftershock’s field log documents the lives-saved headline we had to walk back. IncidentGym’s log records scoring functions that did not work before the ones that did. This honesty is the point: it is the only way to build systems that other people can trust.
What comes next
The next phase moves from hackathon laboratories to enterprise-scale applications. The patterns I have developed in these public builds — deterministic simulation, auditable scoring, multi-agent coordination under pressure — are the same patterns that enterprise organisations need for operational resilience.
The builds continue. The questions get bigger. The laboratories get more rigorous.
The best way to understand a system is to build it. The best way to trust a system is to break it in public and show what survived.