/systems/designed · built · handed over
Production AI, with governance built in
Plenty of people use AI tools. I design production systems: architectures that run daily, survive handover to non-technical operators, and hold up in regulated environments. Everything on this page has run in production, not in a demo.
A content engine that runs without me
Built freelance for CJ Digital, a Melbourne agency producing content for more than 50 clients across seven verticals. The problem was economic: quality content production doesn't scale on human hours alone, and cheap automation produces the generic sludge that Google and readers both punish.
The system handles topic discovery and writing end to end. Each client has a voice profile, so the output reads like them, not like a model. Every draft passes through two rounds of critical self-review and a factual verification pass before a human sees it. A self-learning loop absorbs changes in SEO and AI-search best practice without manual prompt updates. And the whole thing was designed for zero-handover operation: the agency's non-technical staff run it day to day, which is the hardest constraint and the one that matters most.
Annual production value recovered for the agency
API cost replacing roughly $1,260 a month of production labour
Market-rate cost of the builds, delivered freelance
A fully autonomous agent
The second CJ Digital system is a Google Business Profile agent: content generation, rotation, scheduling and publishing across 21 clients with zero human input, running on a cloud server for about $30 a month. I defined the use cases, designed the evaluation frameworks, and set the quality boundaries it operates inside. It's a small system with a large lesson. Full autonomy is an architecture decision you earn through constraint design, not a switch you flip. The agent can only act within boundaries that make its worst output acceptable, which is exactly how autonomy should be scoped in any organisation that answers to a regulator.
AI adoption inside a big four bank
At ANZ I led the structured adoption of AI across the content function: capability building, guardrails, and workflow redesign rather than tool distribution. The result was a 50% improvement in production velocity with zero compliance incidents. That second number is the one I lead with in regulated environments, because velocity without a clean compliance record is just risk with better branding.
Rankline, my own product
Rankline is a content engine for Australian truck and equipment finance brokers: hub-and-spoke page systems, blog articles, LinkedIn posts, one-pagers and newsletters, built in the broker's voice with verified finance facts, for one fixed monthly fee. It's built on the same thesis as this site: broker content now has three readers, and the third one, AI assistants deciding which brokers to name, is where a growing share of enquiries will be decided.
I run Rankline outside my employed work, in a market deliberately distant from it. It's where I test the method in public, with my own money on the line. rankline.ai →
What I bring to an AI engagement
The rare combination is not prompt skill. It's systems architecture plus regulated-industry judgement: knowing where the human gate belongs, what the audit trail needs to show, and how to design for the operator who inherits the system after you leave. If you're weighing up an AI content or discoverability initiative and need it to survive both the compliance review and the handover, that's the conversation I'm built for.