AI with a concrete task.

From assessment to application.

My experience includes production AI tools with Microsoft Foundry and vision models, among them the structured analysis of meter images. What matters to me is the whole application: what data is available, how results are checked and how people continue working with them?

• Assessing concrete AI use cases and technical feasibility
• AI agents as tools, for example for transforming and preparing data
• Tools and interfaces that AI agents can use
• Planning and implementing LLM and vision integrations
• Bots, classification and handoff to people
• Pipelines with clearly scoped tasks for parallel agents
• Local and self-hosted AI with models such as Devstral or Qwen
• Personal reviews of AI-generated code and its integration
• Tool selection and mentoring for development teams

Using AI deliberately. Saving time and effort.

I use AI to analyse large logs and codebases faster and to build applications more quickly. What matters to me is the economic benefit across the whole development process. That takes careful preparation, planning and technical know-how: I develop the concept, define the architecture and code structure and prepare clearly scoped tasks with the necessary context. I choose models and tools to suit the task. This avoids unnecessary model calls, token costs and correction loops. AI and agents support analysis, code and reviews; I integrate and check the results myself. I also teach this workflow to development teams. I work with Mistral, Claude and OpenAI as well as locally run models such as Devstral and Qwen. Provider, operating model, licence, hardware and data flows are matched to your requirements.