ELISA, a requirements ELIcitation Smart Agent — Comparison of state-of-the-art LLMs

Despite the growing adoption of Natural Language Processing (NLP) in requirements engineering, few recent studies have focused on its application in the early phase of requirements elicitation. This paper presents a method to assess the ability of three state-of-the-art large language models (LLMs), Mistral NeMo, GPT-5 and Qwen3-Max to conduct interviews aimed at eliciting client needs, system requirements and environmental pre-conditions, and producing a requirements document. To this end, we developed ELISA (ELIcitation Smart AI-based Agent) designed to support requirements elicitation through an iterative process of question generation and answer collection, followed by the automatic synthesis of the collected information into a requirements document. ELISA is parameterized by a modular connection to multiple LLMs, enabling flexible and comparative evaluation. The study examines ELISA interview generation and requirements reporting capabilities from verification and validation perspectives across four projects and three LLMs.