AI Expertise
Artificial intelligence that delivers in practice.
AI is not a future topic for me and not a buzzword. It is a daily tool. I have used it productively since its first public day and help companies introduce it where it creates real value: as agentic systems integrated deep into your workflows, with frontier-level performance and GDPR-compliant. Because those who bring AI into their processes early and enable their teams secure a real edge in efficiency and competitiveness.
The difference from many AI consultants: I come from software development and from project management. I build AI-powered workflows and agents myself, not just on slides. And I know the difference between what works in production and what merely looks good in a demo.
My Background
At the Forefront from Day One
I studied computer science and bring a deep technical understanding. When the first ChatGPT went public, I immediately recognized how disruptive this technology is, and that you have to get on board early to avoid being left behind.
From the very start I stocked up massively on knowledge to play at the very front. Artificial intelligence is an extremely fast-moving field: new models, new tools, and new hardware all the time. Whoever falls behind here rarely catches up.
Today I am right at the front of this field: whether cloud automation via AI agents, agentic workflows, or connecting entire system landscapes to AI. I don't follow the development from a distance, I work with it every day.
From Practice
Company-Wide AI Rollout
In a client engagement I introduced AI company-wide and guided the rollout of Claude and Claude Code across the entire organization. I connected the existing systems to the AI and optimized the workflows so that work processes run significantly faster and more efficiently.
I am not just an adviser here, but a daily user myself: whatever I introduce, I first prove on my own work.
When AI Support Pays Off
Does This Sound Familiar?
You're stuck in pilot mode.
First AI experiments are running, but nothing reaches production. This is exactly where most AI initiatives stall. You need someone who turns pilots into a viable operating model.
Your processes are ripe for automation.
Reporting, documentation, research, and the breaks between your tools: much of it can be reliably automated today with AI agents. You just don't know where to start.
You want frontier performance, but GDPR-compliant.
Your data must not flow uncontrolled to US providers. There are solutions that combine top performance with data protection. They just have to be set up correctly.
Your team should master AI, not just use it.
Tools alone are not enough. I enable your employees and leaders to use AI and agent harnesses correctly and productively in their daily work.
What I Do for You
AI Services at a Glance
AI Strategy & Use-Case Assessment
I analyze your processes and find the points with the greatest leverage. You get no technology study, but a prioritized roadmap with concrete use cases, clearly aimed at the path from pilot to production.
Agentic AI & Multi-Agent Systems
I design and build autonomous AI agents that research, review, prioritize, document, and act. Orchestrated via standards like the Model Context Protocol (MCP), with clear guardrails and human-in-the-loop where it matters.
Workflow Automation & System Integration
I connect AI to your existing systems, from ERP and CRM to internal tools, and build automated process chains across system boundaries. Value comes from deep integration, not from a chatbot on top.
Sovereign, GDPR-Compliant AI
I select the right model for each use case and design the architecture around it: which model, where it runs, and how your data flows. From an EU-hosted frontier model to your own open-weight model in your own data center, fine-tuned to your task where needed. Performance and data sovereignty in the right balance, not an off-the-shelf default.
Coding Agents & Agentic Software Development
I roll out coding agents like Claude Code into development teams. Developers move from writing code to orchestrating it, with tangible productivity gains and no loss of quality.
Training & Enablement
I enable your employees and leaders to operate AI and agent harnesses cleanly and effectively. Hands-on, on real tasks, with the tools that actually work. So AI does not stay with individuals but delivers maximum impact across the whole organization.
Sovereignty & Competitiveness
A Competitive Edge, No Compromise on Data Protection
Large companies adopt AI markedly slower than agile mid-sized firms. The reason is rarely the technology, but the concern that data flows to US providers. That concern is valid, but solvable.
GDPR-Compliant and Still Strong
Via EU data residency, data-processing agreements, and zero data retention you use frontier models without giving up control of your data. For maximum sovereignty, self-hosted open-weight models are added.
Those Who Wait Fall Behind
You stay competitive by bringing AI into your processes early and broadly, and by giving your employees access to these tools. That noticeably raises the efficiency and performance of your teams. Those who hesitate end up slower and more expensive than the competition.
The Hurdle Is Organizational
Legacy systems, governance, and procurement slow things down, not the models. This is exactly where I come in: from an honest assessment to an anchored operating model with clear guardrails.
My Approach
AI Rollout in 5 Steps
Assessment
Where does your organization stand, which processes are suitable, which data is available? Honestly evaluated, without tech romance.
Prioritize use cases
Not everything at once. We start where the value is greatest and the risk is smallest.
Pilot
One real use case, implemented and measurable. Proof instead of presentation.
Rollout into operation
What works gets rolled out: anchored in processes, connected to existing systems, built for daily use.
Enablement & Anchoring
Training, guardrails, and governance so the solution lasts beyond my engagement.
Assessment
Where does your organization stand, which processes are suitable, which data is available? Honestly evaluated, without tech romance.
Prioritize use cases
Not everything at once. We start where the value is greatest and the risk is smallest.
Pilot
One real use case, implemented and measurable. Proof instead of presentation.
Rollout into operation
What works gets rolled out: anchored in processes, connected to existing systems, built for daily use.
Enablement & Anchoring
Training, guardrails, and governance so the solution lasts beyond my engagement.
Model Customization
Only as much customization as needed
Not every task needs its own model. For each task I pick the right model, from a frontier model via API to a local open-weight model, and then work step by step: the simplest lever first, measured on real tasks. Only when the measurement shows it falls short does the next step follow. At the end of the ladder sits fine-tuning, for exactly those tasks where nothing else carries.
Prompt and context engineering
Clear instructions, the right information in context, and a defined output format. With current models this solves a large share of tasks, quickly and without any training effort.
Move on when: the model lacks a way of working or access to your systems.
Skills and tools
If the model lacks procedural knowledge, it gets Agent Skills with working instructions and checking routines, plus access to your systems through the Model Context Protocol. It loads only what the task needs at that moment.
Move on when: knowledge is needed that is current or proprietary.
RAG: knowledge at runtime
Documents are searched on every request and handed to the model, always at their current state. Hybrid search, reranking, and a test set of real questions mean retrieval accuracy is measured rather than assumed.
Move on when: format, tone, or speed are off despite good retrieval.
Fine-tuning for specific tasks
If a concrete bottleneck remains, I train a model for the task itself: for a fixed output format, short response times, lower cost per request, or data that must not leave the house. Through supervised fine-tuning, reinforcement fine-tuning when results can be checked automatically, and distillation when a large model should pass its skill on to a small one. Efficient through LoRA adapters, with QLoRA also on compressed (quantized) models, local or as a managed service.
The effort: Building training data, measuring results, and maintaining the model as new base models appear. That is why I use fine-tuning deliberately and not as a default.
Every step is measured.
Before the first change there is a test set of real tasks, and the same test runs before and after every further step. The result decides, not enthusiasm for technology. Dataset and evals are what lasts: they carry over to the next base model when the market moves again.
Tools & Tech Stack
What I Work With
From frontier models and coding agents to self-hosted open-weight models, including the leading Chinese models. I pick the right tool for each use case.
Agents & Orchestration
Coding Agents & CLIs
Frontier Models
Open-Weight Models (Self-Hostable)
Model Customization & Retrieval
Sovereignty & Deployment
Certifications & Training
Grounded, Not Improvised
Since 2023, continuous, hands-on training in agentic AI, LLM-based automation, and modern AI tools.
Building AI Agents & Automated Workflows (n8n + MCP)
AI-Driven Project & Taskforce Management
Applied AI Coding with GitHub Copilot, Claude Code and ChatGPT Codex
Developing Skills, Plugins, Hooks, MCPs & CLIs for Technical Project Workflows
Advanced Prompt Engineering for Technical Leaders
Efficient Use of AI Harnesses
Contact
Where does AI create real impact in your organization?
Let's clarify that in a short conversation. No obligation, fully confidential. Often 30 minutes is enough to see where the biggest leverage lies.