
Client under NDA (logistics and defense) · Information systems for logistics and defense
AI cataloging in a closed environment without the cloud
For a defense-systems vendor we designed a multi-agent architecture on local LLMs and proved it with a PoC on real documentation. No cloud involved.
Project facts
Every engagement rests on a few checkable facts.
- Our role
- Strategic partner: workshop, architecture, PoC
- Industry
- Information systems for logistics and defense
- Technology
- Local LLMs (on-premise) · Multi-agent architecture · Proof of concept
- Services
- AI agents & automation
A catalog where every technical detail matters
The client builds specialized information systems for logistics and defense. Its solutions run inside critical infrastructure, where accuracy and security are not negotiable. In its cataloging system it maintains extensive product catalogs in which every technical detail matters.
Cataloging new products depended on deep expert knowledge. Specialists went through unstructured documentation by hand: PDF manuals, spreadsheets, technical drawings, images and handwritten notes. The work was slow and error-prone.
The client wanted to replace manual extraction of technical parameters, run powerful AI in a closed on-premise environment and reduce the risk of errors in critical data.
No data leaves the building
- All data must stay in a closed environment
- No cloud AI services
- Language models must run locally
- Cataloged data is critical; errors are not acceptable
A workshop, an architecture and proof on real data
- An expert workshop: how to select, test and operate local language models and how to build specialized AI agents in a secured environment
- A multi-agent architecture concept: how agents cooperate on automatic data extraction and how they assist experts in classifying products
- A targeted proof of concept on the client’s real data
A local model reads complex documentation without the cloud
The proof of concept ran on real documentation. A well-chosen local model extracts key information even from complex technical documents. Feasibility is confirmed, without a single call to the cloud.
The client now has an architecture concept, the know-how from the workshop and a verified base on which it can develop cataloging with local models further. The path holds under conditions the client cannot change.
Similar work, a different starting point
A description in your own words is enough to start.
Describe what is holding you back. Within two business days we get back to you with a first practical step.


