The rapid advancement in medical imaging techniques has generated a wealth of data that can be exploited for diagnostic and treatment planning. Automatic segmentation of brain tumors from Magnetic Resonance Imaging scans is a critical step for precise diagnosis and monitoring. This paper presents an end-to-end pipeline for brain tumor segmentation using 3D UNet and feature extraction using a 3D Autoencoder. The proposed model segments tumors into different classes, including necrotic and non-enhancing tumor core, peritumoral edema, and GD-enhancing tumor. Additionally, the model predicts patient age and survival days using the Support Vector Regression algorithm on latent features generated by the Autoencoder. Experimental results show promising performance in terms of segmentation accuracy and predictive validity.