This project presents the development of a web-based application for brain tumor classification and segmentation using deep learning. The system utilizes a two-stage model pipeline: a ResNet50-based classifier to detect tumor types (glioma, meningioma, pituitary, or no tumor), and an Attention U-Net with a ResNet34 encoder for tumor region segmentation. The classifier achieved over 96% validation accuracy, while the segmentation model obtained a Dice Similarity Coefficient (DSC) of 0.7838 and a Jaccard Index of 0.7077, demonstrating effective performance. The application interface was developed using Gradio, offering a straightforward three-step flow: image upload, loading feedback, and result presentation. The segmentation step is conditionally triggered only when a tumor is detected, improving efficiency. Comparative evaluation shows that integrating attention gates and a pretrained encoder significantly improves segmentation accuracy. Functional testing confirmed the system operates as intended, offering a seamless and informative experience to users. This project demonstrates the feasibility of combining classification and segmentation in a single diagnostic workflow and highlights the potential for AI-powered tools to assist in early tumor detection through intuitive web platforms.