Learning AI by working on concrete problems

Uninform commissioned Yempik to deliver Module 3, “LLMs and Generative AI,” in its AI for Business Processes master’s programme. The module ran online across four days, from 8 to 11 June 2026, with seven participants outside IT roles.

The teaching question was practical: how do you move from familiarity with AI tools to recognising where they can help a process, and where human judgement is still needed?

Caffè Brevia, a fictional roastery created for the master’s programme, provided the through-line. One shared set of documents and problems let participants explore generative AI, retrieving information from sources, and designing solutions while carrying context from one day to the next.

Short lectures alternated with demonstrations and labs. Teams compared tools, worked through a document set with conflicting information, and built chatbot prototypes in Claude Code on the final day. Each participant developed a concept connected to one of their own processes.

The advanced document-retrieval lab generated more questions than planned. It did not fit in its original form, so part of it was revisited in a shorter version on day four. That is a useful limitation to share: dense work takes time, and compressing it too far comes at a cost.

For later editions, we introduced a parking lot for questions that need individual follow-up and left more room for complex labs. When a topic is new, training quality also depends on the space people get to try, fail, and correct their work.

The module did not promise a certification or company-wide transformation. Its output was a first considered project: each participant connected what they had learned to a concrete activity and had the option to present their concept to the group.

That is the standard we use for applied training: people should leave with a better question about their own process and an idea they can assess, test, and develop.