AI, pedagogy, and the future of learning

Agile teams document requirements as user stories — short, user-perspective sentences. On a real project there are hundreds of them, and they are often the only documentation the team has. Past a certain volume, no one can hold the system, its features, and their dependencies in their head. A literature review and interviews with agile practitioners confirmed the pattern: the recurring failures cluster around communication and shared understanding, not around coding.

Conceptual models solve that problem in other settings, but drawing them by hand costs exactly the effort agile teams refuse to spend. So the dissertation removed the cost. I designed an agent-based framework that interprets a set of related user stories written as Behavior-Driven Development (BDD) scenarios, converts them into a structured intermediate representation, and applies a set of mapping rules — implemented as algorithms — to produce four conceptual models automatically: a domain model, a use case model, a process model, and a state machine.

A natural language processing tool automates the whole text-to-model path: feed it BDD scenarios, get the four models back. The approach and the tool were then evaluated through interviews with agile experts. The conclusion: the documentation burden that keeps models out of agile projects is an artifact of manual effort, and once generation is automated, the models earn their place.

The instinct behind this work is the same one that drives my classroom research — find where practice actually breaks down, then build the thing that fixes it and test it with the people who live in that practice.