Abstract: High-fidelity computational fluid dynamics (CFD) methods for resolving the multiscale features of turbulent flows are playing an increasingly critical role in the design and pre-research of aerospace and advanced engineering systems. However, in complex engineering flow scenarios, high-fidelity approaches such as large-eddy simulation (LES) often suffer from poor robustness and strong dependence of the predictions on numerical schemes and model parameters. To address this challenge, the present study proposes a novel framework that integrates data-driven artificial intelligence with LES. Within this framework, intelligent agents are introduced to regulate the CFD solver, thereby ensuring that key predicted flow quantities remain consistent with physical principles or experimental measurements. The framework preserves the mathematical rigor of the CFD solver while enhancing its applicability to specific complex flow scenarios. Through two representative application cases, we demonstrate the superior performance of this data–intelligence-enhanced simulation approach. The results highlight not only its capability for extrapolation to flows of similar nature but also its ability, under prescribed data/physics constraints, to capture dominant flow structures with higher accuracy. Moreover, the approach naturally inherits the interpretability of intelligent learning techniques.