Glioblastoma is an aggressive brain tumor requiring accurate monitoring of treatment response. This work addresses the 2025 BraTS Tumor Progression Challenge task of classifying glioblastoma treatment response using longitudinal magnetic resonance imaging scans based on standardized assessment criteria. We propose RECAP-Net, an end-to-end deep learning pipeline combining spectral-normalized generative adversarial network-based augmentation, Swin UNETR based custom segmentation and an ensemble of three-dimensional convolutional neural network architectures (ResNet, DenseNet, EfficientNet). Multi-modal pre- and post-treatment Magnetic Resonance Imaging volumes and segmentation masks are preprocessed and fed into the model. Our approach captures tumor progression over time and handles class imbalance effectively. Experimental results demonstrate high accuracy and robustness, supporting its utility for automated treatment assessment in clinical neuro-oncology.