Pancreatic cancer (PC) remains a therapeutic challenge due to intrinsic radioresistance, underscoring the urgent need for predictive biomarkers to stratify patients for personalized radiotherapy. Here, we integrated transcriptomic data from the Gene Expression Omnibus (GEO) database (4 radioresistant [RR] and 11 radiosensitive [RS] patients) and 178 pancreatic adenocarcinoma (PAAD) samples from TCGA to identify hub genes governing radiotherapy response and develop a machine learning-based prognostic model. A random forest survival analysis identified three immune-related hub genes (CARMIL2, APBB1, and ADAMTS13) significantly associated with overall survival (OS) and key features of the tumor immune microenvironment (TIME). The Cox regression-derived risk model stratified patients into distinct high- and low-risk groups (log-rank P < 0.001), demonstrating robust predictive accuracy (5-year AUC: 0.83). Low-risk patients exhibited enriched immune activation pathways (e.g., CGAS-STING and JAK-STAT), while high-risk patients showed increased sensitivity to Epirubicin, Ruxolitinib and other drugs. Immunohistochemistry (IHC) confirmed the differential protein expression of the hub genes in PAAD tissues. This study establishes a translational framework linking multi-omics data and immune modulation to radiotherapy sensitivity, offering actionable biomarkers (CARMIL2, APBB1, ADAMTS13) and a prognostic tool for personalized therapy. Prospective validation and functional studies are warranted to advance clinical application.