Sign language is the primary mode of communication for the deaf community, but hearing people's unfamiliarity with it creates significant barriers to accessing essential services and educational settings. This thesis tackles these issues by developing Ethiopian Sign Language (ESL) recognition technology. Current ESL recognition systems have limitations in terms of accuracy, computational efficiency, and real-time performance, making it difficult to integrate them into everyday life via edge devices such as mobile phones. The purpose of this study is to address these issues by improving accuracy and ensuring real-time responsiveness using deep learning and computer vision algorithms. The data collection process included gathering and preprocessing a dataset of 8400 images representing ten well-known letters used by Ethiopia's hearing-impaired people. A comparative analysis method is used in this study to compare how well three prominent deep learning models MobileNet V2, Faster-RCNN, and YOLO recognize letters in Ethiopian sign language. The results show that SSD MobileNet V2 outperformed Faster-RCNN and YOLO with an accuracy of 94.76%. Furthermore, the development of a mobile application to detect and classify Ethiopian sign language gestures demonstrates not only the research's practical application in improving communication accessibility for hearing-impaired people, but also a significant step toward more assistive technologies for the Ethiopian hearing-impaired community. Keywords: Sign language, Deaf community, ESL recognition technology, Deep learning, Accuracy, Real-time performance.