Efficient requirements engineering is crucial to building a robust software system. Traditional approaches such as stakeholder interviews, surveys, and document analysis often produce incomplete, ambiguous, or inconsistent requirements, leading to design issues. To overcome these challenges, this study investigates the use of large language models (LLMs) to automate and improve the process of requirements gathering and use case generation. LLMs offer advanced natural language understanding features that enable them to process large volumes of textual data, identify inconsistencies, and generate comprehensive and well-structured requirements. In addition, paraphrasing is used to generate diverse user story variations, enhancing the model’s ability to understand different sentence structures and reducing ambiguity during requirement analysis. Two LLMs, BART-base and Falcon 7B, are finetuned to meet domain-specific requirements, improving their capability to generate well-defined and structured requirement specifications in JSON format. In this study, a comparative evaluation is conducted to assess their performance in reducing ambiguity and improving the completeness of the requirements.