Stop 04 · AI Foundry
And this is me building with AI, hands-on.
I don't sponsor AI work from a distance — I build with it. Enterprise agentic AI running over live match data on one side of the street, and things I've made myself on the other: a tram tracker, a live transcription app, a basketball game, a shop run by AI agents.
Building it myself is how I can tell a real capability from a demo, and a real estimate from a hopeful one.
AI Foundry · The Forge
Judgment born of actually building
Hands-on with AI, not sponsoring it from a distance
The most valuable product people are the ones who know what AI is good at, what it isn't, how to avoid the pitfalls, and how to write the prompt that actually gets the outcome.
AI is a solution. What matters more is the problem you're pointing it at. AI belongs in both halves of the craft: in how products get designed, and in what the product does for the user. I build with it hands-on, which is how I can tell a real capability from a demo.
Hands-onProblem selectionAI-assisted building
AI Foundry · The Agent Hive
AFL × Microsoft — live match-data agentic AI
Enterprise flagship · Azure · agentic AI over live data
The flagship enterprise AI initiative in my current portfolio: agentic AI working over live match data during games. I was hands-on in the build and mentored teams through internal workshops — not a sponsor watching from a distance.
Consumer-facing AI is one of the six pillars of the product strategy I authored and took to board endorsement.
Agentic AILive dataAzureWorkshop mentoring
AI Foundry · The Signal Room
Agents that work while I sleep
Personal systems · multi-agent orchestration
A set of AI agents that continuously scan the job market, filter roles against my criteria and surface what's genuinely worth a look. Agent orchestration applied to a real problem I actually have — which is the only way I trust a pattern before recommending it at work.
Designing what each agent owns, and where humans stay in the loop, is the whole discipline.
Multi-agentScheduled runsFiltering & ranking
AI Foundry · The Court
Hoops4
Personal build · hardware meets screen
A prototype arcade game bridging physical and digital: sensors in basketball hoops driving a Connect Four-style two-player competition on a TV screen. Hardware input, real-time game state, screen output — designed and built end-to-end.
Sensors / IoTReal-timeGame logic
AI Foundry · The Booth
Live transcription → AI commentary
Personal build · streaming speech-to-text + generation
An app that transcribes live audio and turns it into an AI commentary feed in real time. Built to test where the latency and quality ceilings actually sit when you push speech-to-text and generation together.
Speech-to-textStreamingLLM generation
AI Foundry · Tram Stop 12
A tram tracker for exactly one person
Personal build · deliberately tiny software
A tram tracker built for exactly one station — mine — and tuned to my actual walking speed. A deliberately tiny piece of software that a general-purpose app could never justify building.
That's the point. AI-assisted coding changes what's worth building at all.
AI-assisted codingTransit APIsPersonalisation
AI Foundry · The Corner Store
An AI-built e-commerce venture
Venture · what two people can run in 2026
Helping a friend get a DTC e-commerce business off the ground — I built the Shopify stack, holding pages and email flows, plus five Slack-based AI agents running day-to-day operational tasks.
A working answer to the question of what a two-person business can actually run on now.
Slack AI agentsShopifyEmail automationOps agents
AI Foundry · The Drafting Office
Roadmap tooling & RICE scoring
Built tool · prioritisation with a spine
A tool I built to customise and communicate the digital product roadmap to executive and external stakeholders. Separately, I used AI to parse an intake form and produce a RICE score for new feature requests — so prioritisation started from a consistent quantitative baseline instead of whoever asked loudest.
Built toolGitHubRICEIntake automation
AI Foundry · The Refinery
ML in production, at platform scale
Recommendations, transcription and personalisation for 1M+ users
Directed AI/ML recommendations and transcription across a platform serving over a million signed-up users — personalisation driven off a first-party data stack on GCP with Salesforce Data 360. Production ML with real consequences, not a pilot.
I've since added the governance experience to take first-party data further — consent management, CDP architecture taken to board approval, and a working command of privacy reform — so the personalisation and the permission to do it are designed together rather than retrofitted.
Recommender systemsTranscriptionGCPCDP