This study leverages the RAVIR dataset, which comprises high-quality retinal images captured using infrared scanning laser ophthalmoscopy (SLO), to improve the segmentation of retinal blood vessels. We introduce a cascading methodology that employs two sequential U-Net models to enhance prediction accuracy. The first model performs binary classification of retina masks. These masks are then concatenated with the input images and fed into the second model. The second model features one encoder and three decoders, designed for segmenting arteries, veins, and reconstructing the input image for regularization purposes. Unlike traditional U-Net approaches, our method introduces a novel sequential design that improves vessel delineation by incorporating mask predictions from the first model. The RavNet architecture significantly outperforms existing U-Net-based models, demonstrating its effectiveness in accurately delineating retinal blood vessels. These findings highlight the clinical potential of our approach, with future work aimed at data augmentation, multimodal imaging, real-time applications, transfer learning, and comprehensive clinical validation.