1Z0-184-25 Question 144
Select 2Your data science team is building a new Retrieval-Augmented Generation (RAG) application using Python in Oracle Cloud Infrastructure (OCI). They need to store large domain-specific documents, generate embeddings for those documents, and retrieve relevant passages to feed into a Python-based generative model in OCI. Which two guidelines must you follow to ensure a scalable, secure, and efficient RAG pipeline in OCI?
- A
Store unstructured documents in OCI Object Storage, create text embeddings using an Oracle-supported or open-source Python model, and maintain embeddings in Oracle NoSQL Database or Autonomous Database with a vector index.
- B
Send all your documents as one large consolidated request to the model each time, relying on the model to find context without any retrieval step.
- C
Use ephemeral storage within Oracle Functions to retain embeddings across all future function invocations and scale automatically.
- D
Use OCI Identity and Access Management (IAM) policies to restrict access to data, ensuring only authorized entities can read from the embedding store and Object Storage.
- E
Deploy an on-premises vector store for embeddings to avoid using any cloud-based database service, which simplifies OCI data security requirements.
Show answer and explanation
Correct answers: A, D
Explanation
Building a RAG application in OCI typically involves storing documents in OCI Object Storage, creating embeddings (e.g., via a Python library running in OCI Data Science or another compute service), and persisting them in a vector-capable database like Oracle NoSQL Database or Autonomous Database. This approach allows efficient similarity searches via vector indexes. Oracle IAM policies should be used to secure access. For more details, refer to the Oracle NoSQL Database documentation on vector search and the OCI Identity and Access Management guides.
- A. Correct.
Correct. Storing documents in OCI Object Storage and creating embeddings with an appropriate Python model is a common pattern in RAG workflows. Oracle NoSQL Database or Autonomous Database can store and index embeddings for fast similarity searches.
- B. Incorrect.
Incorrect. Sending all documents as one large request without retrieval is inefficient and defeats the purpose of RAG. The model� generation step should use only the most relevant data, which you retrieve using embeddings and vector search.
- C. Incorrect.
Incorrect. Oracle Functions provides ephemeral storage that is erased after each invocation, so it� not suitable for storing embeddings. You need a persistent, scalable store for embeddings and indexes.
- D. Correct.
Correct. Applying IAM policies to limit read or write access to stored documents and embeddings is essential for data security in OCI.
- E. Incorrect.
Incorrect. Using an on-premises vector store detaches your pipeline from OCI� native services, complicates integration, and generally increases latency. OCI offers managed services like Autonomous Database or NoSQL Database for vector data storage.