The long-term behavior of dissipative dynamical systems often evolves on low-dimensional manifolds embedded within the high-dimensional phase space. In chaotic dynamics, these solution manifolds can exhibit intricate geometric structures, making it challenging to accurately represent them with a single reduced-order model (ROM). In this work, we propose a framework that constructs a collection of local ROMs, each of which tailored to capture the dynamics of specific regions of the manifold. To achieve this, we quantize the manifold using cluster-based analysis. Within each cluster, the fluctuation dynamics relative to the centroid are modeled using intrusive Galerkin Proper Orthogonal Decomposition (POD) ROMs centered at the corresponding centroid. Transitions between the ROMs of different clusters are governed by an assignment function. Our methodology is verified on the Kuramato-Sivashinsky equations and applied to the Kolmogorov flow across different flow regimes. The proposed framework offers a computationally efficient and scalable strategy for real-time prediction and control of turbulent flows.