For smart grids that are integrating more renewable energy sources, accurate load forecasting and good demand side management (DSM) are very important for keeping them stable, efficient, and long-lasting. This study presents an AI-driven framework that integrates sophisticated forecasting models-Extreme Gradient Boosting (XGBoost), Long Short-Term Memory (LSTM), and a streamlined Transformer-with an optimization engine for Demand Side Management (DSM). We preprocessed historical smart meter data, weather variables, and tariff signals and used them to train forecasting models. When we looked at performance over several time periods, we found that the Transformer always did better than baseline models, with lower RMSE and MAPE values, especially in multi-step forecasting. The forecasting results were then put into a DSM optimization module, which planned when to charge electric vehicles, use storage, and schedule flexible loads. The results show that DSM not only lowered the peak-to-average ratio, but it also saved a lot of money without making the users less comfortable. Sensitivity analysis validated resilience in the context of dynamic pricing and renewable variability. The proposed framework shows how AI-based forecasting and DSM optimization can work together to make future power systems more scalable and easier for consumers to use.