Rapid diagnosis is essential for efficient management and care Alzheimer's disease a degenerative neurological ailment presents a major barrier in terms of early identification and severity categorization. This study uses Grad- CAM for interpretability and ensemble learning approaches to address the urgent need for precise and effective identification of Alzheimer's severity. To improve model performance and generalization the dataset was extensively preprocessed using MRI scans, comprising data augmentation, standardization and feature scaling. Inception V3, ResNet50 and VGG19 were among the deep learning models used they all achieved noteworthy train-test accuracies: Inception V3 (98.94% train, 90.94% test), ResNet50 (95.60% train, 79.55% test) and VGG19 (99.42% train, 93.14% test). For real-time detection the VGG19 model was chosen because of its exceptional accuracy and resilience. Heatmaps were created using Grad-CAM, which successfully highlighted areas of fascination with MRI pictures and made forecasts visually interpretable. A user-friendly streamlit interface was developed to assist in the real-time diagnosis of Alzheimer's severity. It was separated into four groups: “Mild Dementia,” “Moderate Dementia,” “No Dementia” and “Very Mild Disease.” The interface provides comprehensive information on the symptoms of each dementia kind as well as preventative measures. Additionally, the system provides each patient with a thorough report that they can download for later use. The individual's name, age, diagnosis and Grad-CAM heatmap representations are all included in this report. This innovative method combines exceptional precision, interpretability and accessibility to provide an effective solution for early Alzheimer's detection and personalized treatment.