AI in production, not on slides

Starting point

There is no shortage of AI promises, but productive AI systems are rare. Most projects die as proofs of concept, because the hard questions live between the demo and daily operations: error cases, integrations, operations, responsibility.

I wanted to prove the opposite at my own expense: an AI product that runs in the daily business of real property managers, with real money attached, EUR 25 million in invoice volume per year.

What I was responsible for

• ✓ Built the product end to end: design, frontend, backend and operations on Azure, as a one-person development effort with AI support
• ✓ AI indexing for three core workflows: mail intake, invoice intake, contract management
• ✓ Built "Lotse", an assistant that answers questions about the document base with sources
• ✓ Implemented six ERP integrations: facilioo, Aareon Karthago and PowerHaus, GFAD haussoft, WinCasa, Impower, plus the DocuWare EU cloud
• ✓ Owned sales, partner contact and customer acquisition myself

From demo to daily operations

The difference between an AI demo and an AI product lies in the unspectacular details: what happens to an invoice the model reads incorrectly? How does a document reach the archive in an audit-proof way? How does the property manager notice that something is stuck? Exactly these details decide whether a system is switched off after four weeks or extended after four months.

This experience feeds directly into the "AI agent in 4 weeks" offer: I know from running my own product what a productive AI workflow needs, and I build it in from the start.

Proof

• liegenwerk.digital

AI indexing DocuWare Azure .NET ERP integrations