The demand for lightweight energy-absorbing structures has rapidly increased. Auxetic structures, such as re-entrant honeycombs, exhibit negative Poisson’s ratio behavior and high impact stability; however, their compressive response is highly nonlinear and sensitive to geometric parameters, making them computationally expensive and design inefficient. In this study, we present an end-to-end machine learning design acceleration framework for a quasi-2D re-entrant honeycomb that integrates PyAnsys Geometry and PyMechanical. The proposed framework employs automated geometry generation and finite element analysis (FEA) to build a dataset (n = 40) forinterpretable surrogate modeling and covariance matrix adaptation evolution strategy optimization. Experimental validation was conducted using printed digital light processing specimens. An ensemble surrogate model (combining polynomial regression and gradient boosting regression) achieved the best 5-fold cross-validated performance (R² = 0.729). Surrogate-based optimization yielded an optimal design that increased energy absorption density (EAD) in refined-mesh FEA by 63.6% relative to a baseline design (EAD: 90.015to 147.304 kJ/m³). Also, compression tests confirm enhance in pre-fracture energy absorption. Evaluated under a strain limit of ε = 0.15 (due to premature fracture of the optimized specimen), EAD increased by 83.96% relative to a baseline design (EAD: 37.96 to 69.83 kJ/m³). The proposed framework provides a validated method for accelerating the optimization of porous structures under nonlinear responses.