Plant diseases reduce crop productivity, making early and accurate detection essential for food security and sustainable agriculture. However, conventional disease identification techniques are frequently time consuming, labor intensive, and remote in rural regions, where fast diagnostic equipment is not available. This paper overcomes these issues by introducing a light-weight and effective deep-learning approach to the diagnosis of cereal crop diseases on devices. Using the MobileNetV2 architecture as part of a transfer learning set-up, the presented model is trained on a hand-curated dataset over 16 classes of diseases on prominent cereal crops such as rice, wheat, maize and sorghum. In an effort to optimize for mobile device deployment, the model is quantized through TensorFlow Lite, resulting in memory and computational savings.. The quantized model is deployed in real time through an Android-based mobile application, enabling disease detection on the device for farmers in rural areas. This research adds a practical and scalable solution to equip farmers with AI-based disease diagnosis in resource-limited agricultural environments.