<p>Identifying the most stable configuration in nanoclusters is a critical challenge due to the complex energy landscapes and numerous metastable states. Conventional optimization techniques are often thwarted by slow convergence and a propensity for premature convergence to suboptimal solutions. To surmount these challenges, we have utilized a sophisticated deep reinforcement learning framework (DRL) to navigate the intricate potential energy surface of the nanoclusters and predicted the most stable, highly symmetric Ag₁₃ (Ih) and Au₇Ag₆ (Ih) configurations. The deep reinforcement learning agent employs atom-centered symmetry functions to encode structural features into a high-dimensional state space, enabling precise representation of atomic configurations. By incorporating trust region policy optimization into this deep reinforcement learning framework, the agent dynamically balances the exploration of uncharted potential energy surface regions and the exploitation of energetically favorable pathways. The framework not only pinpoints the most stable structure but also explores a diverse array of local minima, offering valuable insights into the structural and energetic attributes of the Ag₁₃ (Ih) and Au₇Ag₆ (Ih) nanoclusters. This validates the applicability and robustness of the DRL framework.</p>