Christian Heilig

Productive AI systems, no prototypes.

I develop AI agents and multi-agent systems that run permanently — with AI-powered development. My role in this is that of an architect and auditor: I define architecture, requirements, and quality rules, guide the models through implementation, and check every result against real evidence — test runs, logs, actual outputs. Nothing is considered finished because it sounds plausible. This is the difference between a demonstration prototype and a system that still does what it's supposed to after six months.

Proven with my own system, in daily continuous operation: voice control with wake word, command center with live data, automatic mailbox evaluation, own accounting (EÜR/USt, cent-accurate, with change lock), 23 background services, over 1,100 automated tests. It was created in audit cycles with multiple AI roles — architect, developer, security and quality auditor — who must refute each other before anything is adopted. I bring this way of working to projects.

What I implement: agents that act independently instead of just responding, with memory, tool access, and traceable decision logic. Multi-level architectures across three tiers, so that responsibilities remain separate and parts remain interchangeable. Process automation in Python instead of modular toolkits: customizable, without license binding and monthly third-party costs. Document and email processing, interfaces to existing systems, RAG connection of own knowledge bases.

Data protection: for sensitive cases, local models and on-premise — the data does not leave your premises. Where cloud models make sense (Claude, OpenAI, Gemini, Grok), they remain switchable depending on the task, instead of tying you to one provider.

Collaboration: remote, project language German, fixed price or daily rate.

AI Agent Development · Multi-Agent Systems · RAG · LLM Integration · Prompt Engineering