Rich-quench-lean (RQL) burners are a promising option for hydrogen applications due to their high flame stability and flashback safety. However, their physical complexity poses a challenge for flamelet-based combustion models, a popular state-of-the-art modelling technique. Stratification, partial premixing, and flow-induced flame stretch create regions where both premixed (P) and non-premixed (NP) burning regimes coexist, giving rise to multi-regime (MR) structures. Due to the synergistic interaction between turbulence and differential diffusion (DD) for hydrogen-rich flames, flamelet manifolds must account for DD and potentially also curvature effects to reliably predict flame-dynamics and emissions. So far, flamelet manifold approaches for partially premixed (PP) flames have not accounted for curvature effects, despite their demonstrated relevance for both P and NP hydrogen flames. In a recent study, we utilized a fully resolved simulation of a turbulent 2D lifted PP hydrogen flame to analyze the impact of curvature and partial premixing on the thermo-chemical state-space. The analysis revealed that curvature-induced variations are essential for capturing species distributions in MR hydrogen flames. Based on this finding, we developed the first flamelet manifold that simultaneously incorporates both curvature and partial premixing effects for hydrogen combustion [1]. A deep neural network (DNN) was used to parametrize the flamelet database and construct the manifold. The new manifold was compared against a standard NP manifold, and one accounting for partial premixing without curvature with the same DNN architecture. The results showed that only the novel approach accurately reproduced the density and reaction rates for the MR conditions. Previous studies have demonstrated the coupling of flamelet manifolds with artificial neural networks [2-4] and their training on time-series data. Building on this work, we extend our current CFD-NN flamelet approach to simulate flames of varying complexity and operating conditions. The application represents a significant advancement towards a unified, physically consistent combustion modelling framework for hydrogen RQL engines.