Ear conditions such as acute otitis media, chronic otitis media, earwax plugs, and myringosclerosis require accurate early screening and treatment to prevent the risk of permanent hearing loss. Previous research has applied the YOLOv8 algorithm on a Raspberry Pi device to detect these conditions, but it still faces the challenge of slow inference speed, reaching 1 second per frame. This study aims to develop an ear disease detection system using the state-of-the art YOLOv8s deep learning algorithm implemented on the NVIDIA Jetson Nano edge computing device. The dataset consists of approximately 2,200 ear endoscopy images in YOLO format, divided into training, validation, and test sets. The system was trained to detect six classes of ear conditions: Acute Otitis Media, Chronic Otitis Media, Earwax Plug, Myringosclerosis, Normal, and Undefined. The model was trained using an optimized CNN architecture from Ultralytics on an NVIDIA RTX 3050 GPU with an input image size of 640x640 pixels. Preliminary results show that the YOLOv8s variant delivers better performance and efficiency compared to YOLOv11n. In final testing using the test dataset, the YOLOv8s model achieved a precision of 99%, a recall of 100%, an mAP50 of 99%, and an mAP50-95 of 93%. When implemented on NVIDIA Jetson Nano hardware, the system is capable of processing video inference from a camera endoscopy at an average frame rate of 15 FPS. The system was also successfully integrated with Flask AI Server, Laravel Web Server, and a MySQL database to support the diagnostic process and the storage of examination results. With this approach, the developed system serves as a responsive, standalone, and efficient decision support system to assist medical personnel in conducting initial screenings for ear diseases at primary care facilities.