A team is adding a similarity-search feature to an application and needs watsonx.data to store and query vector embeddings alongside its existing lakehouse tables. Which service should the administrator add in Infrastructure Manager?

The correct answer is the Milvus service, which provides vector storage and similarity search within the platform. Infrastructure Manager manages services as well as engines, and Milvus is the vector database service in watsonx.data. It stores embeddings and serves similarity search using indexes designed for nearest-neighbour queries, which is the workload the application requires. An additional Presto engine, configuring one of its catalogs to hold the embeddings, is incorrect because a Presto engine executes SQL over catalog tables. It can select numeric columns, but it has no vector index and no similarity operator, so a nearest-neighbour lookup would degrade into a full scan and a distance computation for every row. A larger object storage bucket, since embeddings are ordinary columnar data, is incorrect because it treats an indexing problem as a capacity problem. Object storage will happily hold the embeddings; nothing about that makes them searchable by similarity. A Spark engine, which performs similarity search as part of its query planning, is incorrect because Spark is a processing engine used for transformation and batch work. It can compute embeddings as part of a pipeline, but it does not provide the persistent vector index the feature depends on. For the exam, know the roles: Presto for interactive SQL, Spark for processing, Milvus for vectors, all added and managed through Infrastructure Manager.