An Approach to Developing a Software Requirements Specification Template for Greenfield Projects Using Artificial Intelligence

A paper explores an approach to developing a template for software requirements specification in Greenfield Software Projects. A technical architecture was proposed and information flows were defined for such an approach. To validate the presented approach, an experiment was conducted using neural network models, specifically BERT for requirement classification and GPT-2 for requirements generation. As a result of the experiment, based on the initial information received from the customer, requirements were generated and classified according to functionality into functional and non-functional categories. Compared to the initial information from the client, the generated requirements were more accurate and complete.