Tomato leaf diseases pose a significant challenge to global agricultural productivity, emphasizing the need for accurate and timely detection methods to reduce crop losses effectively. This research introduces a lightweight and efficient framework that utilizes MobileNetV2, a lightweight deep learning model, to detect diseases in tomatoes and estimate the severity of the disease. Explainable AI (XAI) is achieved using Grad-CAM visualizations, which highlight disease-affected regions on leaf images, providing actionable insights for farmers and fostering trust in AI-driven solutions. Achieving an inference time of just 10ms on edge devices, the framework outperforms traditional architectures like ResNet50 and InceptionV3 in efficiency, making it ideal for real-time edge device deployment in precision agriculture.