1Z0-184-25 Question 133
Single answerYou are helping your organization build a Retrieval-Augmented Generation (RAG) solution on Oracle Cloud Infrastructure (OCI) to answer customer support queries. The team has about 100,000 technical support documents stored in OCI Object Storage and needs to provide relevant excerpts to a large language model. Which approach will best allow the system to efficiently retrieve and include only the most pertinent information for each question?
- A
Embed all documents as a single vector stored in memory and feed that vector to the model during each request.
- B
Chunk the documents and store their embeddings in a vector-enabled Oracle Database or other vector store, then retrieve the top matches at query time.
- C
Transfer all documents to a third-party tool outside OCI and rely on an external database to perform the retrieval.
- D
Stream the entire Object Storage bucket to the model and rely on the model to locate relevant sections on-the-fly.
Show answer and explanation
Correct answer: B
Explanation
Real-world RAG solutions often store document embeddings in a vector-enabled repository (such as an Oracle Database that supports vector capabilities) for efficient similarity search. This enables quick retrieval of context�usually just the few most relevant chunks�before passing them to a Language Model for further processing. Oracle documentation recommends storing and indexing domain-specific data to reduce retrieval time and improve quality of generated responses.
- A. Incorrect.
Option 1 is incorrect because embedding all documents as one single vector is impractical for large-scale repositories and does not allow granular retrieval. It would be computationally expensive to feed a massive vector at each request.
- B. Correct.
Option 2 is correct. In RAG-based architectures, chunking documents and storing their embeddings in a vector store (which can be an Oracle Database offering vector capabilities) allows you to retrieve only the most relevant chunks. This approach optimizes resource usage and improves accuracy by providing targeted context to the model.
- C. Incorrect.
Option 3 is incorrect because relying solely on an external data store outside OCI may introduce additional overhead and latency. While it could work, it does not leverage an integrated OCI-based vector store or Oracle Database for seamless retrieval.
- D. Incorrect.
Option 4 is incorrect because streaming the entire bucket of documents for every request would be highly inefficient. Large language models perform better when provided concise, relevant context rather than sifting through entire document sets.