Soft skill evaluation through LLMs

This thesis was motivated by an inquiry into the feasibility of using large language models (LLMs) for soft skills assessment. I was initially introduced to XOOTS, a recruitment automation company focused on evaluating hard skills and offering basic soft skills assessments through simple prompt-based questions. This exposure highlighted a critical gap: while current methods may effectively assess technical competencies, they often overlook the complex, multifaceted nature of soft skills such as leadership, communication, and emotional intelligence. Through my collaboration with XOOTS and independent research, I realized that assessing soft skills requires more than static questions. Traditional methods can be easily manipulated or fail to elicit genuine responses, as candidates may provide socially desirable answers instead of revealing their true abilities. Motivated by these limitations, I developed an alternative approach: a dynamic, interactive conversational system that adapts in real time to user responses, simulating realistic scenarios and incorporating adaptive dialogues, emotional state tracking, and multi-modal interaction. Thus, this thesis presents a novel framework for soft skills evaluation by integrating state-of-the-art LLMs with interactive narrative simulations.