Stop Prompting, Start Engineering: How to Get Reliable Models Out of AI

Ask an AI for a SysML v2 model and you get something that looks right. That is the problem.

It parses, it reads well, and it quietly decomposes the system wrong, drops a requirement, or points a flow the wrong way.
The fix is not a better prompt. It is everything you put around the model: a validator that runs on every write, project rules the AI must follow, concept documents it loads before it writes, a semantic check that catches what the syntax check cannot, and a human gate at the end of each step.

That assembly is called a harness. Research on agent harnesses reports up to tenfold reliability gains from changes to the setup alone, without touching the model, and MBSE is an unusually good fit: SysML v2 tools give us a real compiler, which many AI domains simply do not have.

In this workshop, we will build together a system model live, with you choosing the system. We walk the SYSMOD steps from problem statement to architecture, and at every step you see both sides: what the AI produces, and what the harness catches before it reaches the model. Errors are part of the plan. When the validator rejects something, we look at why, and we turn the correction into a rule that the setup enforces from then on.

You leave knowing which parts of an AI setup do the real work, where syntactic validity stops being evidence of anything, and how to build the same guardrails in your own environment.

Format: live modeling session with open questions throughout. No prior AI experience needed. Familiarity with SysML or MBSE helps but is not required.
This session will be held in English.

With Hamza Basssam and Axel Scheithauer (both trainers & consultants at oose)

***
🔎Want to learn more?
At oose.de, you'll find seminars, training, and fresh insights on software development, systems and software architecture, requirements engineering, soft skills, agile ways of working, and new ways of working.
👉 Follow us on LinkedIn too – we regularly share knowledge, events, and a look behind the scenes at oose.