CAR-T cell therapy has demonstrated substantial efficacy in B-cell malignancies, but treatment responses remain heterogeneous, and T-cell exhaustion and functional dysfunction can limit therapeutic outcomes. Understanding the molecular programs that regulate CAR-T cell activation, cytotoxicity, memory formation, and exhaustion is essential for improving therapeutic efficacy. This project aims to characterize and computationally model CD8 T-cell functional states using single-cell RNA sequencing (scRNA-seq) datasets from CAR-T cell studies and relevant publicly available reference datasets. Particular emphasis will be placed on identifying transcriptional differences between functional and dysfunctional CD8 T-cell populations, including exhausted and cytotoxic T-cell subsets. We will integrate conventional single-cell transcriptomic analysis with GPU-accelerated machine learning and pretrained single-cell foundation models. The analyses will include dataset integration, cell-state annotation, differential gene expression, pathway analysis, latent representation learning, and exploratory in silico gene perturbation. These approaches will be used to investigate transcriptional programs associated with T-cell dysfunction and prioritize candidate regulatory genes for future experimental validation. The proposed work will primarily employ pretrained models rather than training large foundation models from scratch, ensuring that the computational workload remains appropriate for a NAISS Small allocation. The expected outcome is a reproducible computational framework for investigating CD8 CAR-T cell heterogeneity, identifying candidate molecular regulators of T-cell dysfunction, and generating testable hypotheses for improving CAR-T cell function. The project will support data-driven research in cancer immunotherapy.