AI Engineering

AI engineering with large language models, mostly open-source models running locally, and cloud models where they fit. I build RAG platforms, agentic systems and LLM workflows end to end: prompt engineering, orchestration, retrieval, evaluation and deployment. Confidential data can stay on self-managed infrastructure.

Cut multi-hour manual processes to minutes with agentic systems that keep humans on the decisions that matter.

Make internal knowledge searchable and permission-aware, with retrieval that respects who is allowed to see what.

Run confidential workloads on local open-source models instead of sending data to a public API.

Start from the real workflow

Map the steps, the decisions, the exceptions and the points where a human must stay in control before choosing a model or a framework.

Engineer around the model

A reliable system is prompt engineering, orchestration, sub-agents, retrieval and deterministic code together. The model alone is never the product.

Deploy and keep it honest

Ship with evaluation, monitoring and clear failure paths, on cloud or on private infrastructure, so quality can be measured rather than assumed.

What have you actually built with LLMs?

A secure RAG platform for a private bank where each department runs its own governed GPT behind one shared chat interface, with authorization checked before retrieval. An agentic recruitment system that took a process from roughly five hours of manual work per case down to about five minutes, using orchestration and sub-agents with humans kept on the key steps. A system that turns a plain-language process description into a valid BPMN 2.0 model. And a Swiss German transcription tool with LLM summaries tailored per department.

Do you work with local models or cloud APIs?

Both, with the focus on local. Most of my work uses open-source models on self-managed infrastructure, which is what makes confidential documents and internal processes workable at all in regulated environments. Cloud models are used where the data allows it and the capability is worth it.

Do agentic systems remove the human entirely?

No, and they should not. In the recruitment system almost everything around the important steps is automated, but a person still approves the decisions that carry consequences. Autonomy is applied where it is safe and withheld where it is not.