Erythropoiesis-stimulating agents (ESAs) are essential for anemia management in hemodialysis patients, but conventional dosing strategies reliant on clinical experience often yield suboptimal outcomes. This study developed artificial intelligence (AI)-driven models to personalize ESA dosing by integrating comprehensive clinical parameters. We retrospectively analyzed 148,381 monthly hemodialysis sessions from 9,738 patients across 8 university hospitals from 2009 to 2022. Two predictive frameworks were compared for prediction of hemoglobin level and ESA dose: a machine learning (ML) model incorporating 25 variables (3 demographic factors [age, smoking history, height], 15 laboratory variables, and 7 dialysis-related measures) with 3-step time lag integration yielding 100 features; and the Deep Learning-Gated Recurrent Unit with Attention mechanism (DL-GRU+ATTENTION) utilizing 16 variables (2 demographic factors [gender, height], 11 laboratory, and 3 dialysis-related measures) also with 3-step time lag integration. For hemoglobin prediction, the XGBoost model achieved mean squared error (MSE) of 0.33, root mean squared error (RMSE) of 0.58, mean absolute error (MAE) of 0.36, and R² of 0.67, while the DL-GRU+ATTENTION model showed comparable error metrics but superior R² of 0.75. For ESA dose prediction, ML models yielded MSEs of 382-431, RMSEs of 10.77-20.00, MAEs of 6.36-10.62, and R² values of 0.003-0.09, whereas the DL-GRU+ATTENTION model demonstrated exceptional precision (MSE: 16.36, RMSE: 4.05, MAE: 1.33, R²: 0.96). The DL-GRU+ATTENTION model significantly outperforms traditional ML methodologies in predicting both hemoglobin levels and optimal ESA dosing. This approach provides a robust foundation for implementing precision medicine in anemia management for hemodialysis patients, potentially improving clinical outcomes.