This research paper explores the application of YOLO-NAS (You Only Look Once Neural Architecture Search) L/M/S in firearm detection within an advanced surveillance system. The increasing incidence of firearm-related crimes necessitates the development of efficient and accurate methods for firearm detection. The proposed solution utilizes the YOLO-NAS L/M/S architecture to enhance the surveillance system's ability to identify firearms in real-time. This paper discusses the architecture of YOLO-NAS L/M/S, its training process, and its performance evaluation in detecting firearms. The results demonstrate the effectiveness of the YOLO-NAS L/M/S approach, highlighting its potential for integration into advanced surveillance systems for improved public safety