1Z0-184-25 Question 64
Single answerYou are designing a recommendation engine within OCI that ranks products based on their similarity to user-generated text embeddings. Your team wants near real-time responses on queries against tens of millions of embeddings stored in Oracle Autonomous Database. Which design approach should you implement to efficiently perform similarity searches with minimal overhead?
- A
Use Oracle Autonomous Database with a specialized vector column type and build an approximate nearest neighbor (ANN) index on the embedding columns.
- B
Store embeddings in OCI Object Storage and run periodic Apache Spark jobs in OCI Data Flow to compute the top similar items for each query.
- C
Store the embedding data as plain text in Oracle Autonomous Database and apply standard text indexing to enable similarity lookups.
- D
Deploy a user-managed GPU cluster outside OCI, replicating Oracle Autonomous Database data nightly, and run a custom vector similarity algorithm on demand.
Show answer and explanation
Correct answer: A
Explanation
For near real-time vector similarity searches on embeddings, leveraging Oracle Autonomous Database� native vector data type and approximate nearest neighbor indexing is the recommended approach. It avoids the latency of batch jobs, is more accurate than full text search for embedding similarity, and reduces the complexity of managing external infrastructure. For more details, consult Oracle� official documentation on vector search and indexing with Oracle Autonomous Database.
- A. Correct.
Correct. Oracle Autonomous Database supports a vector data type and indexing approach designed specifically for similarity searches using embeddings. By storing embeddings in a vector column and creating an approximate nearest neighbor index, you gain efficient and near real-time similarity lookups. This setup aligns with the requirement to handle tens of millions of embeddings with minimal overhead.
- B. Incorrect.
Incorrect. While OCI Data Flow and Spark jobs can be useful for batch analytics, they introduce significant latency between runs. This is not suitable for near real-time similarity lookups because the recommendation engine must wait until the Spark job completes to refresh its similarity results.
- C. Incorrect.
Incorrect. Storing embedding data as plain text and using a text index is not optimized for vector similarity searches. Traditional text indexing is designed for keyword-based retrieval rather than computing cosine similarity or other distance metrics on numerical embeddings.
- D. Incorrect.
Incorrect. Hosting GPUs outside of OCI introduces additional complexity and latency due to data replication and network overhead. While a GPU cluster can accelerate vector operations, the question specifically requests minimal overhead and near real-time responses within OCI, making this approach less optimal.