
Enterprise AI stack, built in-house
Four GPU hosts, one model family for every chat lane, automatic failover and an honest look at five weeks of rebuilding.
A cross-section of applications, LLM systems and infrastructure, built and in production.
These references show production work: no mockups, no concept studies. They're organized around how we actually operate: "01 · Apps" covers web and enterprise applications that replace or connect processes; "02 · LLM" covers our own AI models on our own hardware, with no detour through third-party APIs; "03 · IT" is the operations underneath, the infrastructure that carries these systems. The collection is growing: eight projects are documented here today, three more are in internal review.
Customer data in copy and images has been neutralised.

Four GPU hosts, one model family for every chat lane, automatic failover and an honest look at five weeks of rebuilding.

Company-wide logs in one place, and only the alerts that matter.

From Excel chaos to a group cockpit: numbers for Monday morning and not month-end.

ChatGPT comfort in-house: 5 models, 32 tools, ~90 user accounts. Chats and documents stay in-house.

26 small watchdogs and data bridges that report before month-end does.

Websites that get found: by Google and by AI search.

55 controls, clear owners, automatic reminders: compliance as a living process.

Seven warehouse workflows on the handheld: from goods-in to the carrier label.