skip to content
~/lab/projects/ireya.mdactive

lumi

active

An internal AI chat wired into a live S/4HANA landscape — the argument being that the model is the commodity and the connection to your own systems is the part nobody else can sell you.

year
2026
stack
SAP BTPSAP AI CoreABAPCloud ConnectorClaude

Read the value brief → — the case for building this, as it gets put in front of management.

Staff at any large enterprise are already using AI every day. They are using ChatGPT, which knows the internet and knows nothing about the business — so getting useful help means pasting internal data into a tool the company doesn’t own. That is happening whether or not anyone approved it.

Lumi is the internal answer: the same plain chat box, but connected to the company’s own systems and running inside its own BTP landscape.

the argument

The pitch is not “we built an AI.” Anyone can buy the same models — Claude and GPT are available to every company on earth, so the model is not the differentiator and saying otherwise gets picked apart in the first five minutes of a management meeting.

What can’t be bought is a live connection to a specific company’s S/4HANA, its ABAP code, its processes. That connection is the moat, and it only exists if someone builds it inside the estate. Framing it that way also changes what the ask is: not budget for a model, budget for owning the entry point to AI inside the company.

what it does

It reads real ABAP code and SAP objects in a live landscape, over Cloud Connector, with data staying inside SAP BTP and SAP AI Core rather than leaving for a public endpoint.

Developers are the use case that’s running. The architecture doesn’t care — an AI chat that connects into real systems points just as easily at finance’s reporting tools or an ops team’s plant systems. One platform each team plugs into beats each team starting from zero on the same problem.

what I’d change

The value brief is deliberately a management document — it argues ownership, risk and cost, and it does not pretend the platform reaches every team today. Three of the four boxes in the platform diagram are the same architecture pointed somewhere new, which is a real claim but an untested one. The honest version of the roadmap is that the second team to connect is the one that proves the pattern generalises — and until someone outside the original landscape connects, that remains an argument rather than evidence.