Optimizing leachate recirculation operations in landfills is critical for maximizing resource recovery while maintaining safe operational conditions. This study presents a practical framework for optimizing leachate injection (LI) parameters using Bayesian neural network (BNN) surrogates of a coupled thermo-hydro-bio-mechanical (CTHBM) model. The framework enables efficient prediction of cumulative methane (CH4) generation and maximum temperature evolution, addressing the computational challenges of large-scale landfill simulations. Data generated by varying crucial input variables in the CTHBM model was used to train two different BNNs for predicting CH4 generation and maximum temperature evolution within landfills. The developed surrogate models were able to accurately predict the CTHBM model data, with an R2 value greater than 0.99 for both CH4 and temperature predictions on test data. Using the developed surrogate models, employing expected improvement in energy efficiency over conventional landfilling (no-recirculation) as the selection criteria, an optimization framework was developed that can be used to maximize both biogas and heat energy recovery from landfills, while maintaining landfill temperatures below 65 °C. The applicability of the developed optimization framework was demonstrated using a case study. The optimal LI parameters identified provided higher energy recovery compared to both extreme scenarios i.e., no-injection and continuous LI at high pressure, while maintaining safe operational temperatures. These findings highlight the effectiveness of machine learning-based surrogate modeling in landfill performance optimization for enhanced energy recovery.