Recent advances in deep learning have surpassed benchmarks once considered unattainable, ushering in unprecedented capabilities in various domains, including online marketplaces. In classified ads platforms, search and recommendation systems face unique challenges, such as interpreting user intentions, handling diverse and sparse data, and adapting to changing inventories. To tackle these complexities, we leverage rich user interaction data to train embedding models that transform search queries and ads into compatible vector spaces, enabling efficient and accurate vector-based search and recommendation. This approach allows us to align user preferences with the most relevant content, enhancing the efficacy of our classified ads platform.