1Z0-184-25 Question 125
Single answerYou are designing a Retrieval-Augmented Generation (RAG) application on Oracle Cloud Infrastructure to answer user questions based on product manuals stored as PDFs in an OCI Object Storage bucket. You plan to chunk the documents, generate vector embeddings for each chunk, and ensure minimal-latency retrieval at query time. Which approach best satisfies these requirements while maintaining security and performance on OCI?
- A
Use OCI Data Flow for a PDF-chunking Spark job, generate embeddings using OCI Data Science, store embeddings in Oracle Database with vector search capabilities, and implement a custom API to perform nearest-neighbor queries at runtime.
- B
Directly upload entire PDF documents into the context window of a Large Language Model, relying on the model to manage retrieval dynamically without any embedding or chunking logic.
- C
Copy PDFs into an on-premises file system, manually chunk the documents, and run queries by streaming each file in its entirety through a custom HPC cluster on OCI at query time.
- D
Leverage Oracle Analytics Cloud to ingest all PDFs and present them in a BI dashboard; rely on the BI dashboard filters as the final retrieval layer for RAG queries.
Show answer and explanation
Correct answer: A
Explanation
Building a RAG application on OCI typically involves three main steps: (1) ingest and process documents, (2) generate vector embeddings, and (3) retrieve and serve relevant chunks to a Large Language Model. By using OCI Data Flow to process PDFs in parallel, OCI Data Science to generate or manage embeddings, and Oracle Database (with vector indexing/search capabilities) to store and search embeddings quickly, you create a secure and highly performant pipeline. Refer to Oracle� documentation on using OCI Data Flow, OCI Data Science, and Database indexing best practices for detailed implementation guidelines.
- A. Correct.
This approach aligns with RAG best practices for both efficiency and security. Using OCI Data Flow to handle large-scale processing and chunking, OCI Data Science to generate embeddings, and Oracle Database with vector search capabilities for low-latency retrieval, provides an end-to-end pipeline that scales well and maintains strong data governance.
- B. Incorrect.
Uploading entire PDFs into the LLM context window is not scalable and can lead to high latency or token overflows. It also disregards best practices for chunking large documents into manageable segments.
- C. Incorrect.
Storing PDFs on-premises and manually chunking them removes the benefits of OCI� managed services and can introduce significant performance and reliability challenges. It also complicates network configuration and security, making retrieval slower and more complex.
- D. Incorrect.
While Oracle Analytics Cloud can visualize data, it is not designed for fine-grained, near real-time vector-based retrieval critical to RAG applications. A BI dashboard alone cannot effectively handle embedding queries or provide nearest-neighbor lookup for relevant chunks.