LLM Use in Early Prototyping
Alma Smajić, Mirjana Radetić-Paić, Darko Etinger · Procedia Computer Science · 2026
Early-stage prototyping is crucial in software design and recent advances in LLMs are reshaping this phase. LLM-based tools bridge gaps between requirements and design, allowing faster iterations and more integrated workflows. Prompt engineering lets multidisciplinary teams to rapidly explore design ideas and assess AI capabilities before full-scale development. This paper shifts authors’ previous pedagogical thesis toward a technical focus; examining the role of LLMs in the early prototyping of an accessible application. The case study, NeuRoam, is a tablet-based educational app prototype for neurodivergent children, aligned with WCAG 2.2 and COGA guidelines. Analysis covers how LLM tools accelerate the prototyping of such inclusive apps while ensuring adherence to accessibility and user requirements. Results from a structured evaluation using 17 LLMs showed that models with stronger reasoning capabilities significantly predicted higher usability ratings (β ≈ 0.823, p = 0.002), with the regression model explaining approximately 55% of score variance (R² ≈ 0.555). LLM feedback identified both critical accessibility issues (e.g., low contrast) and praised features such as gamification, highlighting their potential as early-stage evaluators in inclusive design.