How we reach a decision before you invest
Our consulting starts with an inventory, not a pitch deck. In the use-case assessment we test every candidate along three axes: business value, data situation, and technical feasibility. A project moves into delivery only when all three hold. We judge data readiness in concrete terms: whether sources are structured, whether access rights are settled, whether the density of evidence is sufficient for citation-grounded RAG. When a foundation is missing, we say so, rather than launching a proof of concept that is set up to fail.
We work vendor-neutral. The build-versus-buy question is answered per case, not by a favourite product. Our reference architecture is the ESTAYA AI Platform with LiteLLM gateway, PageIndex, and knowledge graph, run on open-weight models such as Llama, Mistral, or Qwen under vLLM. Where standard software is enough, we recommend buying it. Where data sovereignty and depth decide the outcome, we build on our own servers in Germany.
We treat compliance as a framework we implement for you: GDPR, the EU AI Act, and BSI IT-Grundschutz. Every use case is mapped to an EU AI Act risk class before any code is written. That removes rework later.
- Use-case assessment by value, data, and feasibility
- Data-readiness review instead of assumptions
- EU AI Act risk classification before the first sprint
- Build-versus-buy recommendation argued per case
- Enablement of your teams to run it themselves