AI implementation and business architecture
Four years advising enterprise clients on where AI actually belongs in their operation — and, more often, where it doesn’t. The work is roadmap, sequencing and integration: which processes change, what the technology has to prove before it’s trusted, and how the deployment aligns with the budget and the KPIs someone is already accountable for.
Typical output: a staged implementation plan with named owners, a working pilot, and the measurement to tell you whether it did anything.
Autonomous and agentic systems
Multi-agent architectures with persistent memory, defined skills, business rules and safety guardrails — not a chatbot bolted onto a form. I design an agent the way you’d write a job description: what it knows, what it’s allowed to do, whose rules it follows, and where it must stop and ask.
Built on current agentic frameworks and multi-model orchestration, with a review layer that checks the system’s own output before it reaches anyone.
Custom software and hardware engineering
When the capability doesn’t exist off the shelf. C#/.NET 8 and WPF for Windows field tooling, Python and Google Apps Script for automation and data pipelines, React and Firebase for web applications. On the hardware side: reverse engineering, PCB design, 3D CAD and printing, CNC.
The point isn’t the technology list. It’s that a problem doesn’t get parked because it crosses the line between software and a physical object.
Programme and operations management
Thirty years of running things that have to work on a schedule: forecasting, budgeting, KPI design, field team performance, inventory protocol, regulatory compliance (Enaon EDA, Greek net metering, GDPR), and the executive reporting that keeps a programme funded.
This is usually the part that decides whether the technical work survives.
Account Take Over
Taking over Google Ads, Merchant Center and analytics accounts from a previous agency or marketer — for businesses whose reporting has stopped meaning anything. The first job is almost always the same: find out whether the conversion data can be trusted before anyone spends against it.
Structured as an audit first — who has access, what the tracking actually measures, an honest baseline, and what the previous setup did — then a phased rebuild where nothing is deleted and nothing new goes live unreviewed.