Endometrial cancer is one of the most common gynecological malignancies, with endometrial cytology playing a crucial role in early detection and diagnosis. The diagnosis is complicated and time-consuming due to hormone-induced morphological changes in cells and the thickness of cell clusters. Although most currently reported AI support systems for cytology are designed for use with digital images from whole-slide imaging (WSI), the digitalization of endometrial cytological slides is challenging due to focusing issues caused by the thickness of cell clusters. Despite the high demand for artificial intelligence (AI)-supported diagnostic systems, studies of endometrial cytodiagnostic support AI systems have been delayed. This study introduces a real-time abnormal cell cluster detection system utilizing You Only Look Once version 5x (YOLOv5x) under a microscope to bypass the use of WSI. We collected 146 preoperative endometrial cytology cases at Nippon Medical School between 2017 and 2023, confirmed by hysterectomy specimens. YOLOv5x was trained using 3,151 images captured with a simple smartphone setup from the cytology slides of 96 cases. For real-time object detection, images were captured using a charge-coupled device (CCD) camera attached to a microscope and fed directly into YOLOv5x. For real-time model evaluation, we adjusted the threshold settings for the cell-cluster and slide levels using 30 new cases. In the final assessment, the AI model's diagnostic performance was compared with that of human evaluators, including pathologists and medical students with varying levels of experience, across 20 new cases. The AI model surpassed human evaluators, achieving an accuracy of 85%, a precision of 82%, and a recall of 90%. Moreover, AI-assisted diagnosis substantially reduced the time taken by human evaluators by 37% and notably enhanced diagnostic metrics, particularly among pathologists and medical students accustomed to using the AI system. Overall, our findings demonstrate that real-time detection with an AI-supported system substantially enhances early cancer detection, integrates smoothly into existing workflows, and is invaluable in resource-constrained settings without the need for expensive specialized devices.