From AI strategy
to measurable results
I build AI strategies, deliver the systems behind them, and guide organizations through adoption – from the first idea to real change in processes and decisions. 20+ years in financial services IT, backed by working projects of my own, not just slide decks.
Projects Get in touchThis site is powered by the Legion AI system
Most AI initiatives never get past the slide deck
I have seen from both sides of the table how IT projects fail – bad contracts, missing specifications, technology chosen for the hype rather than for the problem. I work differently.
The typical corporate approach to AI
- Deployments that look impressive in the demo and miss the mark in production
- Company data and documents flowing into third-party clouds without full control
- Lock-in to a single SaaS platform, with an ever-growing bill for tokens
- “We bought our people licenses, so the company has AI.”
How I work
- I test and run every solution myself before I recommend it
- Local‑First / Private‑by‑Design architecture wherever data confidentiality matters
- Open, replaceable components (Ollama, Semantic Kernel, open-source LLMs) instead of closed platforms
- Engineering oversight of every line of code – including the lines generated by AI
About me
For more than 20 years I have worked in financial services IT – as an engineering team leader, systems architect, and the person accountable for technology strategy. Today I focus on multi‑agent AI systems, digital transformation, and practical deployments that have to work under regulatory constraints, not only in the lab.
I do not just design – I code. Legion (an autonomous agent network that builds applications from specification through to tests), a local personal-data detection system, the PhoneBot voice assistant integrated with a VoIP exchange, and Edge AI that detects car theft attempts – these are not PowerPoint decks but working solutions I built from scratch and documented step by step.
I translate IT language into business language and back again. A strong analytical background lets me quickly pinpoint the processes genuinely worth optimizing with AI – not because it is fashionable, but because it will deliver a measurable result.
I drive AI adoption across organizations – making sure a rollout does not end the moment the licenses are bought, but genuinely changes how teams work day to day and delivers lasting results.
I build AI adoption strategies as part of a business transformation – not as a standalone IT project. The strategy covers technology, governance, capability building, and outcome measurement. Success depends on close collaboration between business and IT, and on combining domain knowledge with what the technology can actually do.
How I work
✓ I build – before I propose a solution, I prove it in practice.
✓ Data privacy – on‑premise and local‑first architectures wherever they make sense.
✓ No needless dependencies – open source and replaceable components instead of closed platforms.
✓ AI Enabler – I translate technology into business decisions and pinpoint the processes ready for AI.
✓ AI adoption – I make sure a rollout does not end with a license purchase but genuinely changes how teams work.
✓ Regulated-sector discipline – I treat GDPR, DORA, and the EU AI Act as the standard, not an obstacle.
What I have built
Working in IT does not have to be boring. A selection of my own projects, each documented publicly step by step – from the architecture through to the lessons learned and the costs.
Legion – a digital company staffed by AI agents
An autonomous network of agents (CEO, Product Manager, Analyst, UI Designer, Developer, Tester) that takes a description of a business need and independently produces the specification, the code, the tests, and the documentation – without a single line of my own code.
See Legion live →Local‑First personal data detection
An audit system that scans large, unstructured data sets for PII entirely offline – Semantic Kernel, Gemma, Ollama, and a local RTX 5090 card, without a single outbound request.
Case study on LinkedIn →PhoneBot – an autonomous voice assistant
A voice bot integrated directly with a VoIP exchange (Asterisk/FreePBX) and built on native audio‑to‑audio processing – without expensive, closed SaaS platforms such as Retell or Bland AI.
Case study on LinkedIn →Edge AI versus car theft
A computer vision system on an NVIDIA Jetson Nano that detects break-in attempts in real time – with behavioral motion analysis, critical zones, and false-alarm minimization.
Case study on LinkedIn →An AI form coach for the gym
A posture monitoring system that pairs a local vision model (real-time analysis) with a cloud model for deeper post-workout analysis and personalized guidance.
Case study on LinkedIn →I built a digital IT company where every employee is an AI agent. I even sent the CEO into the GPU layer.
I documented the entire delivery process in task.md, pdlc.md, and tech.md, and handed the conductor’s baton for the agent orchestra (CEO, Product Manager, Analyst, UI Designer, Developer, Tester) to a Product Manager enforcing quality gates between phases – with no involvement from me in the code.
Case study on LinkedIn →An agent for optimizing IT infrastructure cost
An AI agent built on Zabbix and Claude Sonnet 5 that analyzes resource utilization every month, spots over-provisioning, and recommends savings – operating strictly read-only, so it never touches anything that protects the system from failure.
Case study on LinkedIn →Reconstructing a system prompt
A security experiment: I set one AI agent against another to see whether a system prompt’s safeguards could be broken. It worked after 68 attempts – the decisive weaknesses turned out to be sentence-completion, content injection through the knowledge base, and translation tricks.
Case study on LinkedIn →ProcessManager – deterministic control over agents
In regulated sectors (finance, insurance), the autonomy of LLM agents alone is not enough. I designed a ProcessManager role that enforces a predictable workflow, quality gates, and human oversight – instead of relying solely on a larger model context window.
Case study on LinkedIn →Copilot wired into company data through MCP
I explored how to connect Microsoft Copilot securely to real business processes through the Model Context Protocol – integrating it with operational data (CRM, ERP), not just with documents. The advantage goes to the companies that can safely wire AI into their own systems, not to those with the best model.
Case study on LinkedIn →An autonomous security audit in 20 minutes
A two-agent system (an architecture orchestrator and a technical analyst running the security tooling) on GLM 5.2 via Ollama Cloud ran the audit on its own, finding a critical vulnerability and several misconfigured services – in roughly 20 minutes and at low cost.
Case study on LinkedIn →How do you bring people back to the office? – AI agents work the problem
I asked a network of six agents with distinct personalities (CEO, psychologist, visionary, naysayer, HR, lawyer) for a return-to-office strategy based on attraction rather than mandate. The result – the office as a premium product, with events, commuting subsidies, and extra leave instead of penalties.
Case study on LinkedIn →A sparring match between two psychopathic AI agents
I put two agents (Claude and Gemini) head to head, negotiating liability for a failed IT project between a software house and a retail chain, testing personality profiles from agreeable through to psychopathic. A heavily profiled agent negotiates coldly and resists pressure, which raises real questions about AI regulation and the future of leadership.
Case study on LinkedIn →A smart aquarium that detects fish aggression
I combined a passion for aquariums with Edge AI: a system on NVIDIA Jetson, MediaPipe, and Google Gemini analyzes fish behavior in real time and detects aggression within the shoal – more proof that neural networks hold up even in hobby projects.
Case study on LinkedIn →Areas of expertise
What I prove out in my own projects, I turn into real value for the teams and organizations I work with.
AI strategy and transformation
I design AI adoption strategies as part of a business transformation – not as an IT project. I cover technology, governance, workforce capability building, and outcome measurement, balancing innovation against security and regulatory compliance.
Multi‑Agent AI Systems
Designing and deploying networks of AI agents that run end-to-end business processes – from requirements analysis through to testing and documentation.
Voice AI and automation
Voice assistants integrated with telephony and CRM/ERP systems, with no need to pay for expensive SaaS platforms.
Edge AI and Computer Vision
Computer vision systems running locally on edge devices – monitoring, security, and automation.
Custom software
Web and mobile applications, ERP/CRM integrations, and systems tailored to a specific business process.
IT consulting and architecture
Technology audits, technical debt reduction, and support for engineering teams in choosing and deploying the right solutions.
Why me
The principles I hold to in every project – my own and commercial alike.
An engineer and practitioner, not a slide-deck specialist
I design and code every solution before I recommend it – so I know exactly what I am proposing, and why.
Regulated-sector discipline
20+ years in finance taught me to stay immune to hype and to treat GDPR, DORA, and the EU AI Act as the standard rather than an obstacle.
Interested in working together?
Email me, call, or connect on LinkedIn – I am always glad to talk about new professional challenges and interesting projects.