Maverick Studios

// writing

Beyond Tokens harveygill.substack.com ↗

Why AI needs world models to reason, plan, and act reliably. Five essays on the architectural shift from token prediction to state modeling — and why the failures we see are structural, not accidental. Each essay ships with a working reference implementation on AWS.

↗ github.com/MaverickHQ/beyond-tokens — Python, AWS Bedrock, plan → simulate → verify → execute

Executable World Models harveygill.substack.com ↗

A research framework for deterministic agent systems. Trajectories, not outputs. Agents execute, constraints enforce, artifacts record every decision — producing structured datasets ready for evaluation and learning. The architecture that makes intelligent behavior reproducible and inspectable.

↗ github.com/MaverickHQ/executable-world-models — Python, agents, constraints, trajectories, evidence policy

// projects

crucible-player-coach github.com/MaverickHQ ↗

active open source Python SQLite Claude API

Adversarial quality loop for trading decisions. A PlayerAgent proposes trades against a formal constraint schema. A CoachAgent evaluates every proposal mechanically — approves or rejects with named violations. The full exchange is recorded as a structured artifact in SQLite. The scorecard is the product.

Tromsø Aurora Hunter github.com/MaverickHQ ↗

static Globe.gl Three.js public APIs

Real-time aurora probability visualiser. Globe.gl + Three.js. Zero server dependency. Public APIs, service worker, GitHub Pages hosted.

// code

MaverickHQ github.com/MaverickHQ ↗

All open-source tooling and releases, paired with the essay series.

// about

$ whoami

Harvey Gill — architect, builder, research essays with code.

I write two essay series on Substack: Beyond Tokens and Executable World Models. Each essay ships with a working GitHub release under MaverickHQ.

I build tools that reason about the world. Clean architecture. Real-world utility. No scope creep.