1Z0-184-25 Question 130
Select 2Your organization is designing a Retrieval-Augmented Generation (RAG) pipeline on Oracle Cloud Infrastructure (OCI) to deliver dynamic, domain-specific answers. You currently store raw documents in OCI Object Storage and generate vector embeddings using Oracle AI Services. You need to ensure minimal latency and high accuracy for your RAG solution. Which two practices best address these requirements?
- A
Store all document embeddings in Oracle Functions� memory to immediately serve retrieval requests
- B
Use Oracle Database 23c with vector search or Oracle NoSQL Database for low-latency vector-based retrieval
- C
Rely exclusively on third-party, unmanaged solutions for indexing and retrieval to minimize infrastructure overhead
- D
Incorporate domain-specific synonyms and terminologies during both embedding generation and query prompting
Show answer and explanation
Correct answers: B, D
Explanation
For a RAG solution on OCI, the best practices include using OCI-native services that support vector indexing (e.g., Oracle Database 23c or Oracle NoSQL Database) to reduce latency and improve retrieval performance. In addition, tailoring embeddings and prompts with domain-specific terminology ensures relevant context, boosting the accuracy of your generated responses. Refer to Oracle documentation on vector search in Oracle Database 23c/NoSQL Database for recommended deployment patterns and performance considerations.
- A. Incorrect.
Option 1: Incorrect. Oracle Functions provide ephemeral runtime environments and are not designed to hold large embeddings in memory. Storing all embeddings in Functions could cause memory and scalability issues.
- B. Correct.
Option 2: Correct. Oracle Database 23c and Oracle NoSQL Database both offer vector search capabilities to store and quickly retrieve embeddings at scale, aligning with OCI best practices for RAG pipelines.
- C. Incorrect.
Option 3: Incorrect. While third-party solutions can be integrated, relying solely on unmanaged services may complicate security and operational overhead. OCI-native solutions are generally recommended and well-supported.
- D. Correct.
Option 4: Correct. Tailoring the embeddings and queries with domain-specific synonyms or specialized lexicons enhances retrieval accuracy, helping the Large Language Model generate more contextually precise answers.