I build secure AI integrations and EU-hosted custom software for small and mid-sized companies, including regulated industries.

Most of my work sits in the same place: a company already runs a stack that works, and wants AI inside it rather than instead of it. That means retrieval over messy internal knowledge, MCP servers that expose existing tools to language models, and automation that leaves the surrounding systems alone.

Based in Villach, Austria. Everything I ship runs on EU infrastructure.

Website · LinkedIn · jukeep · neurit.ai

• Retrieval and knowledge systems. Vector search, hybrid retrieval, and citation-backed answers over internal documents. The hard part is rarely the embedding model, it is what counts as a source and how you prove an answer.
• MCP servers. Exposing company data and internal tools to LLM clients in a way that is scoped, auditable, and boring to operate.
• Custom software. Next.js, Postgres with pgvector, Docker, self-hosted on Hetzner and netcup. No US-only dependencies in the critical path.
• Process automation. Multi-provider model routing, cost per operation as a hard constraint, and local models wherever a frontier model is overkill.

• jukeep — knowledge memory for companies. Captures decisions and context, answers questions with sources attached instead of plausible-sounding text.
• neurit.ai — lateral thinking platform. Six thinking modes that deliberately return unexpected answers to break a fixed frame.
• yoorda — QR ordering and POS for restaurants.

• eigenlag — computes the max-plus eigenvalue of a recurring data pipeline: the shortest cycle time it can sustain without falling behind. For Airflow, Dagster and dbt schedules.
• schemawerk — modern visualisation for CMI heating schemas on Technische Alternative UVR16x2 controllers.
• denkdisziplin — a thinking framework as a Claude skill. Six gates a claim has to pass before it counts as a conclusion.

• Email: info@webse.at
• Web: webse.at