Accurate demand forecasting is critical for modern retail supply chains to ensure optimal inventory management and reduce operational inefficiencies such as stockouts and overstocking. This paper presents an end-to-end cloud-native machine learning architecture for daily store-level retail demand forecasting. The proposed system integrates Amazon Web Services (AWS) components including Amazon S3 for scalable data storage, Amazon Athena for serverless analytics, SageMaker Feature Store for consistent feature management, XGBoost for predictive modeling, and SageMaker Model Monitor for production monitoring. The pipeline performs data ingestion, feature engineering, model training, batch prediction, real-time deployment, and automated monitoring. Experimental evaluation demonstrates the effectiveness of gradient boosting models combined with engineered time-series features for forecasting retail demand. The architecture highlights how cloud-based MLOps practices enable scalable and reliable forecasting systems in production environments.