1Z0-184-25 Question 93
Select 2Your retail company needs to build a near real-time product recommendation engine using vector embeddings in Oracle Cloud Infrastructure (OCI). You want to store and query embeddings for similarity searches (e.g., top-K most similar products) with minimal latency and without manual overhead. Which two approaches can you realistically implement?
- A
Leverage Oracle Autonomous Database with built-in vector indexing capabilities and use similarity-based SQL queries for top-K retrieval
- B
Store all embedding dimensions as individual numeric columns in a relational table, then manually implement B-tree indexes on each dimension
- C
Persist embeddings in OCI Object Storage and run approximate nearest neighbor searches directly on the stored objects
- D
Deploy a specialized approximate nearest neighbor framework (e.g., FAISS) in a container on OCI, maintain the index on persistent storage, and expose a low-latency API for similarity queries
Show answer and explanation
Correct answers: A, D
Explanation
To perform real-time vector similarity searches in OCI, either use Oracle Autonomous Database� built-in vector indexing and search features or deploy a specialized library (e.g., FAISS) in a containerized environment. Storing vectors in numeric columns with manual indexes is typically inefficient for high-dimensional data, and Object Storage on its own does not provide vector search. Refer to Oracle Database documentation on vector-enabled indexes or OCI Container documentation on hosting specialized inference frameworks for best practices.
- A. Correct.
Correct. Oracle Autonomous Database supports vector data types and specialized indexing, providing efficient similarity search for real-time queries. This eliminates much of the manual complexity of building and maintaining custom indexes.
- B. Incorrect.
Incorrect. While you could store embeddings in numeric columns, a purely manual approach with B-tree or other generic indexes is cumbersome and inefficient, especially for high-dimensional data. This leads to poor performance compared to native vector indexing.
- C. Incorrect.
Incorrect. Object Storage is ideal for storing files and objects but does not natively support vector similarity queries. You would have to download and process vectors externally, incurring higher latency and complexity.
- D. Correct.
Correct. Hosting an approximate nearest neighbor index in a container or VM on OCI enables you to query embeddings in near real-time. You would manage the index (e.g., FAISS) and store it in a persistent volume, then expose an API for the recommendation engine. This approach can perform well for large-scale vector data.