Virtual Stakeholder with Generative AI: A Practical Approach to Teaching Requirements Elicitation
Requirements Engineering, Requirements Elicitation, Generative Artificial Intelligence, Large Language Models, Design Science Research.
The teaching of Requirements Engineering faces significant challenges, particularly in the elicitation phase, in which students have limited opportunities to practice real interviews with stakeholders, restricting the development of essential competencies. This work presents the Virtual Stakeholder, an educational tool based on Generative Artificial Intelligence designed to support this process. Developed using Large Language Models and the Retrieval-Augmented Generation technique, the solution simulates realistic interviews with clients, encouraging the use of investigative strategies and effective communication. The research, conducted following the Design Science Research approach, included an evaluation with 35 undergraduate students in Software Engineering, based on qualitative analysis of the interactions and quantitative analysis of the elicited requirements. The results indicate evidence of student engagement and the development of practical competencies related to requirements elicitation, although technical limitations were observed. The study contributes to addressing gaps in the practical teaching of Requirements Engineering and points to directions for future improvements and validations.