Plant diseases remain a persistent challenge to global agriculture, especially in regions with limited access to diagnostic infrastructure and expert support. While deep learning has shown promise in image-based classification tasks, its real-world application in farming remains underdeveloped. This study presents a lightweight, web-based deep learning model for early plant disease detection using RGB images of affected leaves. The proposed model is built upon the MobileNetV2 architecture and fine-tuned using transfer learning to balance accuracy and computational efficiency. Crucially, we sourced and curated a robust, diverse dataset from open-access agricultural repositories and academic archives. This dataset comprises 54 disease classes across 26 plant species, chosen to reflect real-world conditions. This realistic approach is essential for training the model to perform accurately on new, unseen images, ensuring it is a reliable and effective tool for farmers in the field. Data augmentation techniques were applied to further enhance model robustness. We also deployed this model in an intuitive web application that allows to users to upload leaf images, receive instant diagnostic results, and view tailored treatment suggestions. This work demonstrates that MobileNetV2, coupled with a carefully designed dataset and accessible interface, can bridge the gap between state-of-the-art AI and field-level agricultural needs, providing a scalable, user-friendly tool for real-time crop disease management and contributing to food security through technological innovation.