Databricks Generative AI Engineer Associate Question 239
Select 3You are designing a Retrieval-Augmented Generation (RAG) application using Databricks. To serve features effectively for the retrieval step, which resources are most critical to include in your architecture?
- A
A feature store to manage and serve vector embeddings
- B
A vector database optimized for similarity search
- C
A high-performing GPU cluster for model inference
- D
Streaming pipelines to update embeddings in real-time
- E
A data lake to store raw unprocessed data
Show answer and explanation
Correct answers: A, B, D
Explanation
In RAG applications, serving features such as vector embeddings is critical for the retrieval step. A feature store ensures efficient management and access to embeddings, while a vector database provides optimized similarity search capabilities. Additionally, streaming pipelines enable real-time updates of embeddings, ensuring the retrieval process remains accurate and relevant. GPUs and data lakes, though important for other tasks, are not directly tied to feature serving for retrieval.
- A. Correct.
A feature store is essential for managing and serving vector embeddings, which are critical for the retrieval step in RAG applications.
- B. Correct.
A vector database is required for similarity search, enabling the application to retrieve the most relevant context based on embeddings.
- C. Incorrect.
While a GPU cluster is important for model inference, it is not directly related to serving features for retrieval in a RAG application.
- D. Correct.
Streaming pipelines are crucial for keeping embeddings up-to-date, ensuring the application retrieves the most relevant and recent information.
- E. Incorrect.
A data lake is useful for storing raw data but does not directly serve features or embeddings for retrieval in a RAG application.