Demand forecasting addresses perishability, seasonality, and demand uncertainty in agricultural supply chains. This paper presents an able, integrated framework that integrates Long Short-Term Memory (LSTM) based accurate demand forecasting models with Reinforcement Learning (RL) to enhance the accuracy and responsiveness of agricultural demand predictions. The LSTM model represents time-varying patterns, and the RL framework minimizes holding, ordering, and shortage costs through adaptive decision-making. The framework is trained on real-world agricultural data, capturing the temporal dependencies and decision-based feedback in supply planning. The proposed model achieves improvement in prediction accuracy, such as Root Mean Square Error (RMSE) and Mean Absolute Error (MAE). Traditional models are static and unable to model dynamic behavior in Agri-markets. This integrated approach supports decision-making across the agricultural supply chain while LSTM captures time dependencies and RL adapts to changing supply conditions. Limitations and future extensions include scalability and integration with real-time weather data.