My research focuses on explainable artificial intelligence (XAI) and deep reinforcement learning (DRL) for the control and optimization of complex dynamical systems, with emphasis on turbulence and aerospace applications. I develop frameworks that embed explainability tools, such as SHAP, into control loops to identify causally important flow structures and design transparent, interpretable control strategies.
On the computational side, I combine DNS and LES simulations with reduced-order models based on variational autoencoders, DeepONets, and graph-based methods. These models create efficient latent spaces where DRL agents can operate, guided by physically meaningful reward structures. I also integrate uncertainty quantification to ensure robustness of surrogate models and policies. By merging physics, interpretability, and learning, my work advances reliable optimization and control strategies for high-dimensional, nonlinear systems.