We propose a vector embedding approach for recommendation systems aimed at identifying product affinities and suggesting complementary items. By capturing relationships between products, the model delivers highly relevant recommendations based on the context. A neural network is trained on purchase data to generate word embeddings, represented as a weight matrix. The resulting model predicts complementary products with top-20 and top-50 precision scores of 0.59251 and 0.29556, respectively. These embeddings effectively identify products likely to be co-purchased, enhancing the relevance and accuracy of the recommendations.