Lifelong Learning in Software Engineering: Towards an AI-driven Tutor for Cloud-Based Software Architectures

The emergence of large language models
(LLM's) like ChatGPT helped to utilize artificial intelligence
(AI) to understand basic concepts of programming
in university courses, but also to automate and assist in
repetitive procedures in the software engineering process.
However, existing approaches to learn the skills and best
practices to create software architectures and the underlying
domain models with the help of AI expect a certain level
of experience in software engineering. Besides, current
approaches to configure LLMs with prompt engineering to
answer as an AI tutor rather assist students in learning
basic programming skills in university courses than in
software architecture design. In this paper, a four-stepapproach
is proposed to prompt LLMs, namely ChatGPT-
4o and Llama 3.1, to act like a tutor and help novices in
software engineering build their own software architectures,
apply best practices and understand them. To do this, the
LLMs receive initial prompts with the assignment of a
tutor role, and descriptions of the target group, system's
requirements and context. Then, example questions, domain
models, and software architectures will be sent to the LLMs
and evaluated if the answers relate to the models and
requirements and comply with the initial prompt. Tests with
the MobSTr dataset showed overall suitability of these LLMs
for tutoring in software design, but also limitations regarding
some general answers and the processing speed of locally built
LLMs.

DOI 10.1109/E-TEMS64751.2025.11239087

Artificial Intelligence

Education

Large Language Models

Software architecture

Teaching