Domain generalization remains a critical challenge in deep learning, where models are required to generalize effectively to unseen domains. While distributionally robust optimization (DRO) has shown promise in addressing this issue, traditional approaches typically have relied on single-perspective risk assessments, limiting their ability to capture complex domain interactions. To address this problem, we propose an enhanced DRO framework that incorporates domain-specific experts to evaluate risks across all domains, thereby expanding the space of worst-case scenarios. By building a shared feature extractor across various domains and domain-specific classifiers, the proposed method ensures comprehensive risk evaluation and robust learning across diverse domains. Empirical results on publicly available benchmarks showed that our method achieves superior generalization performance under complex domain distribution shifts, outperforming traditional DRO techniques. This work highlights the potential of multi-perspective risk assessments in improving domain generalization performance.