Abstract:

Students in data science courses often struggle to move from a broad interest to a focused, workable project direction. EduIDEAtor is a web-based “first-mile” assistant that guides students through a short ideation flow and returns compact project suggestion cards. This paper reports a formative, design-oriented analysis of student improvement suggestions and translates them into implementable interface requirements for the next iteration and classroom deployment. Using an anonymised post-use survey (N=16), we conducted an inductive thematic analysis of the optional open-ended improvement prompt (n=6) and triangulated priorities with ordinal quantitative diagnostics. Five improvement directions emerged: recoverability (back navigation), richer presentation beyond plain text, output transparency (explaining metrics and topic sections), steerable literature recommendations (broaden, summarise, justify relevance), and consistency of end-step options. A “keep” signal highlighted the importance of preserving low friction. We synthesised these findings into a prioritised design backlog and an evaluation plan for a follow-up classroom study using behavioural outcomes beyond self-report.

Keywords:

Generative AI, large language models, data science education, project-based learning.