Implementing efficient product recommendation systems constitutes a persistent challenge for small-scale e-commerce enterprises. Insufficient computational resources and sparse user-item interaction data often precipitate suboptimal recommendation accuracy. In this study, we address these challenges through several complementary methodologies: (Sect. 3.1) a modified collaborative filtering algorithm that is optimized for sparse interaction matrices and demonstrates superior performance in low data density environments; (Sect. 3.2) a novel data augmentation framework that synthesizes additional training instances based on latent user behavioral patterns and product attribute correlations; (Sect. 3.3) a transfer learning methodology that leverages pertinent knowledge from previously trained large-scale, open-source recommendation models and adapts it to domain-specific applications; and (Sect. 3.4) algorithmic implementations that are computationally lightweight and designed to operate within constrained resource environments. Despite significant resource limitations, empirical tests on e-commerce datasets from different small-scale shops show that our suggested methods operate satisfactorily. Notably, our approach performs noticeably better than traditional recommendation algorithms in terms of user happiness, recall, and precision measures under similar computing constraints. These results imply that, without having to make significant infrastructure investments, small e-commerce businesses can improve their competitive positioning against larger market incumbents by putting advanced recommendation capabilities into place.