Human activity recognition plays a crucial role in various automated systems, particularly in monitoring setups within smart homes to assist elderly individuals. Developing such systems presents certain challenges, including ensuring privacy protection, minimizing costs, and enhancing comfort for individuals residing in the smart home. The developed system should also be designed for flexibility and ease of interaction, ensuring quick response times and enabling prompt decision-making. Recent research has increasingly focused on utilizing Vision-Language Models (VLMs) to develop more flexible and user-friendly interaction systems. However, existing implementations primarily rely on vision sensors, raising concerns about privacy violations. In this study, we address these challenges by presenting a technique for human activity recognition that leverages the Contrastive Language-Image Pretraining (CLIP) model alongside data acquired from three ultra-wideband (UWB) radars. The results demonstrate strong performance, indicating significant potential for practical implementation.