This study aimed to develop a Deep Learning (DL) tool to predict binding between T-Cell Receptors (TCRs) and peptide-bound Major Histocompatibility Complexes (MHCs). Previous studies highlighted the advantages of DL over simpler models, particularly when incorporating BLOSUM-alignment scores and FoldX-derived protein energy terms. However, analysis revealed that BLOSUM scores biased predictions toward negative complexes, leading to overfitting. To address this, BLOSUM scores were replaced with TCR-pMHC amino acid sequences and structural features, significantly reducing bias. A Convolutional Neural Network (CNN), inspired by state-of-the-art protein interaction models, was developed in PyTorch models trained on combined features outperformed amino acid sequences alone in most dataset partitions, with energy terms showing strong predictive performance. Integrating all features enhanced recall of positive complexes, including those missed by sequence-only models.