Algolia's LookingSimilar technology uses machine learning to generate image-based recommendations, unlocking opportunities for visually-driven product suggestions that transcend traditional metadata limitations. By analyzing images and extracting relevant visual features, this approach enables users to discover products with similar aesthetics, even if they can't articulate why they're drawn to them. This is achieved through the use of neural hashes, which produce efficient and cost-effective embedding vectors for recommendation purposes. The technology also incorporates object-level metadata filtering and rules-based boosting to ensure relevance and accuracy. With LookingSimilar, Algolia aims to expand the horizon of visual recommendations, providing a more dynamic merchandising experience that focuses on business needs and rules.