One of the most complex challenges in ensuring software quality is assuring the convergence of the developers' and users' views. Requirements Engineering studies how this can be achieved by investigating how software requirements can be collected and maintained. Nevertheless, it is still unclear how development teams can take advantage of the large amounts of user data found on various social media, app store reviews, and support channels. In this study, we aim to develop and empirically investigate a Machine Learning-powered tool called DeeperMatcher, tailored for agile software development teams to use crowd-based requirements engineering (CrowdRE) to aid the management of their issues and tasks. The research unfolds across three objectives: (I) developing a reliable, maintainable, and organized ML-enabled system; (II) leveraging advancements in natural language processing to provide an approach for CrowdRE; and (III) applying empirical research methods for system validation. Methodologically, we will incorporate a single-case mechanism experiment with a real-world dataset from a specific project developed by our research group and observational case studies from different projects. Our execution plan comprises two phases, the first emphasizing the tool development and validation and the second dedicated to extensive testing and in-depth analysis. Amidst our expected outcomes, we list a well-structured AI-powered system, noteworthy contributions to software and requirements engineering, and valuable insights into the evolving landscape of machine learning in software development.