The development of 5G networks creates complicated demands for effective resource allocation due to varying traffic patterns and a wide range of service requirements. We suggest a hybrid AI model (Harmonia) that blends Transformer designs with Long Short-Term Memory (LSTM) networks in order to overcome these difficulties. This model offers a reliable way to anticipate resource allocation by capturing both short-term and long-term interdependence in network traffic. To improve prediction accuracy, our method makes use of the Transformer’s attention mechanism and the LSTM’s sequence processing capabilities. According to evaluations, the hybrid model outperforms conventional allocation techniques, achieving an R-squared value of 0.5195 and a Root Mean Squared Error (RMSE) of 0.0852. These findings show how the model can adjust to changing network conditions, suggesting that it could be a useful tool for resource management in next-generation mobile network.