1Z0-184-25 Question 120
Select 2You are designing a retrieval-augmented generation (RAG) application to help your company� customer service team quickly respond to user queries based on internal knowledge base documents. These documents are stored as text files in Oracle Object Storage. Your plan involves generating embeddings for these text files, then searching the most relevant passages to embed in LLM prompts. Which two steps are essential when creating this RAG pipeline on Oracle Cloud Infrastructure?
- A
Use Oracle Data Science to generate embeddings from the text data and store these vectors in an Autonomous Database that supports vector search capabilities.
- B
Upload the raw text directly to Oracle Application Performance Monitoring for vector indexing and subsequent retrieval.
- C
Chunk the text documents into smaller sections before generating embeddings to improve retrieval quality.
- D
Create a restrictive IAM policy that denies all network access to Oracle Data Science notebooks to ensure data privacy.
- E
Enable streaming ingestion from Oracle Object Storage to Oracle Streaming Service to generate embeddings in real time.
Show answer and explanation
Correct answers: A, C
Explanation
Building a successful RAG application in Oracle Cloud Infrastructure involves creating embeddings for your text data and storing these vectors in a database capable of vector search. Oracle Data Science allows you to programmatically generate embeddings, and Oracle Autonomous Database (or Oracle Database 23c) provides vector indexing capabilities. Chunking documents into smaller segments greatly improves retrieval precision by matching user queries to more targeted text passages. Refer to Oracle Data Science documentation and the Oracle Database vector search documentation for guidelines on properly configuring and securing your RAG pipeline.
- A. Correct.
Storing vectors in an Autonomous Database (or a compatible Oracle Database with vector search) after generating embeddings in Oracle Data Science is a recommended approach. This lets you perform efficient similarity searches for retrieval-augmented generation.
- B. Incorrect.
Oracle Application Performance Monitoring is not designed for storing or indexing text embeddings. It is intended for monitoring and diagnosing application performance and is not part of a RAG pipeline.
- C. Correct.
Chunking text into smaller sections before embedding is a recognized best practice. It enables more precise retrieval of relevant passages, especially when dealing with large documents.
- D. Incorrect.
Although controlling access is crucial, completely denying all network access to Oracle Data Science notebooks generally makes it impossible to retrieve or process data. You must set IAM policies with the right balance of security and usability.
- E. Incorrect.
Streaming ingestion is typically not required for a foundational RAG pipeline. You usually generate and store embeddings on demand or on a schedule rather than continuously streaming new documents, unless you have a real-time data ingestion scenario.