AI Development Demos
A portfolio built over six months of weekend R&D — real systems, real costs, independently audited. Every one is evidence for a single business question: organisations are pouring spend into AI; is it paying off?
The bill is climbing. Here's what the spend actually returns.
Concrete numbers from the systems below — what a unit of AI work costs, measured rather than guessed.
The bill is only half the equation. The other half is judgment — spend becomes value only when someone can tell good output from bad. That's the discipline this portfolio is built to demonstrate.
Costs in AUD ≈ USD list price × 1.42 (e.g. MiMo US$0.36 · Fable US$51.41). Full per-model data on the 8-model page.
The AI Software Team NEW · FABLE 5
A single one-line brief becomes a reviewed, tested, performance-optimised product — 11 roles (PO, scrum master, BA, architect, UX, engineer, DevOps, QA, security, performance, tech lead) plus five parallel code-reviewers, every one a real Fable 5 agent. Then open the working app it built, live.
Same Brief, 8 Models COMPARISON
The exact same pipeline, run by Fable 5, GPT-5.5, Gemini, GLM, DeepSeek, Kimi, MiMo and MiniMax. Eight apps from one sentence — each tagged with its real cost, security/perf scorecard, and a blunt ✓ works / ✗ broken verdict. Click any to open it live.
When AI Games Its Own Metrics THE AUDIT
The heart of the talk. I audited two of these systems and both had quietly learned to game their own scores — the trading platform ranked money-losers as its "most robust" picks; the newsroom slipped fabrications past its own truth-gates. Every average looked flawless. This is where AI games, and how you catch it.
The Presentation
The keynote: the AI-spend business problem, the intern mental model, where AI games its own metrics, the live 15-role platform build, the two cost dials (model + harness), the governance map, and a path your team can start Monday. Arrow keys to navigate.
AI Trading System
Genetic algorithm engine that evolves options trading strategies across 25 specialised islands. Multi-layered audit system with 283 probes protecting every data path.
Content Pipeline
Fully automated AI content pipeline from research to WordPress. Multi-model orchestration with fact-checking, quality gates, and dual cross-validation scoring.
How I Build
The AI-assisted engineering workflow: how specification, verification, orchestration, and judgment combine — the discipline that turns fast AI output into systems you can actually trust.