Abstract
The translation of plain-language business requirements into precise technical specifications is a cornerstone of successful software development yet remains a labor-intensive and error-prone process. This study explores the application of advanced natural language processing (NLP) models, specifically GPT-3.5, to address this challenge. By fine-tuning the model on a curated dataset of business requirements and corresponding technical specifications, we developed an intelligent system capable of generating structured outputs such as user stories and functional requirements. Our system demonstrated a promising ability to streamline the requirements engineering process, achieving an average BLEU score of 0.72 and garnering positive qualitative feedback from software professionals, who rated the outputs as clear, actionable, and closely aligned with standard practices. However, error analysis revealed occasional over-generation of details and minor omissions, highlighting areas for refinement. This research emphasizes the potential of NLP technologies to bridge the gap between non-technical stakeholders and development teams, reducing the manual workload and facilitating efficient project planning. While further improvements are necessary to enhance accuracy and reliability, our findings mark a significant step toward integrating NLP into practical requirements engineering workflows. Future work will explore expanding the system's capabilities and incorporating iterative feedback mechanisms to achieve greater precision and adaptability.