1Z0-184-25 Question 49
Single answerYou are designing an AI-driven semantic search system on Oracle Cloud Infrastructure using Oracle Autonomous Database. Your application stores text embeddings in a dedicated column, but queries for similar documents are running too slowly. You want to improve the speed of these vector-based similarity searches while preserving efficient performance for standard SQL queries in the same environment. Which approach should you implement to achieve optimal performance for both vector-based and standard SQL queries?
- A
Create an approximate nearest neighbor (ANN) vector index on the embeddings column and configure the index to only be used by queries that require vector similarity search
- B
Replace the existing table with an external file in Object Storage, then scan the file for vector similarities in the application layer
- C
Create a standard B-tree index on the embeddings column to allow both equality lookups and approximate similarity queries
- D
Store embeddings as JSON documents and rely on a JSON text index to perform semantic searches
Show answer and explanation
Correct answer: A
Explanation
To speed up AI vector searches, you should use an approximate nearest neighbor (ANN) vector index on the embeddings column. This specialized index structure allows rapid similarity queries while still enabling efficient standard SQL lookups via conventional indexes. According to Oracle best practices and documentation on vector search in Autonomous Database (introduced in Oracle Database 23c and enhanced in subsequent releases), ANN indexing is the recommended method to handle large-scale embeddings for AI/ML workloads. Refer to the Oracle Database Vector Search documentation for detailed implementation steps and performance considerations.
- A. Correct.
Option 1 is correct because approximate nearest neighbor (ANN) vector indexes are designed to perform fast similarity searches on high-dimensional data. By configuring the index for vector queries only, you ensure regular SQL queries can still use existing relational indexes, preserving their performance.
- B. Incorrect.
Option 2 is incorrect because moving embeddings into Object Storage and scanning externally would significantly degrade real-time performance. This approach might also introduce complex maintenance overhead and is not efficient for high-frequency queries.
- C. Incorrect.
Option 3 is incorrect because B-tree indexes are not suitable for multi-dimensional proximity searches. They are optimal for exact matches or range queries, but not for computing similarity on vectors.
- D. Incorrect.
Option 4 is incorrect because JSON indexing is geared toward document-based queries, not specialized vector proximity calculations. While JSON text indexes can handle unstructured data, they are not optimized for vector similarity search.