Work

Products I've shaped and agents I've built, told as the decision behind them rather than the tech.

Trackunit
Construction tech platform

The problem: Customers already work in AI assistants. Fleet and compliance data locked inside one app doesn't follow them there.
The decision: Open the platform to AI through MCP connectors, so customers can build reports and workflows with the tools they already use — and set a clear boundary for what AI may do on its own and what needs a human.

Trackunit
Access management & safety

The problem: Qualification and certificate rules differ by country, and in practice by customer, because each customer's own requirements sit on top of the legislation.
The decision: Build operator profiles as configuration customers layer their own rules onto, validate training at the source through the IPAF integration, and take Inspections from discovery on building sites through to launch.

Own project
Agent

The problem: Job alerts are noisy, duplicated and often wrong about location or seniority.
The decision: An agent that runs twice a week, verifies every posting at the company's own careers page, scores it on five axes — and reports what it filtered out, so the filter can be calibrated.
What I learned: Neither source can be trusted alone, and they fail in opposite directions. The product work is in the verification rules, not the scraping.

Own project
This site

The problem: A CV answers the questions I chose. Recruiters have their own.
The decision: An openly labelled AI that answers in my voice from approved material only — no salary, no confidential details, no commitments — and a log of what people actually ask.
What I learned: The log is the product: it shows how the market reads me.

Want the story behind one of these? Ask my AI, or read how I work.