Microwave-assisted pyrolysis (MAP) facilitates rapid volumetric heating and high efficiency. However, it is directed by nonlinear and complex microwave-feedstock material interfaces that challenge reliable prediction. In this work, pyrolysis of Sargassum longifolium macroalgae (MA) and waste tyre (WT) was conducted using used lithium ion battery anode material as a microwave absorber material and potassium hydroxide as a catalyst. Equal blending ratio of MA and WT were processed at a microwave power of 300 W and pyrolysis reaction time 600 s, while catalyst and susceptor loadings were varied between 5 and 15 g by addition of 5 g under inert conditions. Improved microwave coupling higher the average heating rate from 18.1 to 60.4 °C/min and enhanced microwave conversion efficiency from 23.6% to 56.7%, shifting pyrolysis product distribution toward high liquid yields (51.7 wt%). In addition, a physics-informed Gaussian Process Regression (PI-GPR) model was developed by applying total mass yield and total specific microwave energy physical conservation constraints. The developed model was achieved high predictive accuracy performance over product yields R2 values in the range of 0.91–0.98. Incorporation of physical constraints reduced ordered mass-balance and energy-balance violations by >99.9%, agreeing to a physics gain (%) more than two orders of magnitude in comparison with unconstrained ML models. Monte Carlo uncertainty propagation yielded narrow 95% confidence intervals and reliable predictive coverage. The proposed architecture determines a strong, physically dependable for optimizing MAP of waste to energy systems.