The rapid expansion of online consumer-to-consumer (C2C) marketplaces has generated a vast amount of image-based data, presenting unique challenges and opportunities for automated analysis. In this work, we propose a robust framework that leverages the DinoV2-base Vision Transformer to convert raw car parts images into high-dimensional embeddings. These embeddings are then effectively reduced via Principal Component Analysis (PCA) and visualized using Uniform Manifold Approximation and Projection (UMAP), revealing intrinsic data structures. By applying k-means clustering and rigorously evaluating cluster quality with the Reduced Silhouette Score, Reduced Calinski-Harabasz Index, and Reduced Davies-Bouldin Index, our experiments demonstrate that a 64-dimensional representation achieves the best balance between intra-cluster cohesion and inter-cluster separation. These promising results underscore the potential of our approach as a scalable and automated solution for monitoring and analyzing visual data in C2C marketplaces.