In geotechnical engineering, the precise identification of essential soil parameters from sensing and experimental data is vital for the accuracy of constitutive and finite element models. However, the complexity of sophisticated soil models often makes this task challenging. Traditional optimization methods that rely on gradient information often fall short in this class of problems, due to their struggle with black box models lacking clear gradient pathways. Gradient-free methods, though circumventing the need for direct gradient data, can still miss out on integrating previous insights when faced with new information. To tackle these issues, our study presents a cutting-edge method inspired by the mechanisms underlying AlphaZero, DeepMind's acclaimed algorithm that excels in mastering complex strategic games through autonomous learning. By adopting a comparable selflearning technique, our approach reinvents the task of parameter identification of advanced geotechnical models as a strategic game. It draws a parallel between optimizing model parameters and the complex task of developing victorious chess tactics. This method utilizes a blend of deep learning for initial estimations and Monte Carlo Tree Search (MCTS) for finer adjustments, promoting a self-enhancing calibration process. Such an approach paves the way for a more self-reliant and intelligent parameter identification methodology from sensing and experimental data. The outcomes of our study demonstrate the robustness and versatility of this approach across various geotechnical models, ranging from the parameter identification of sophisticated constitutive models to more complex applications involving inverse analyses using finite element models that include interactions between mechanical sensing devices and unsaturated soils.