Databricks Generative AI Engineer Associate Question 240
Select 3You are designing a Retrieval-Augmented Generation (RAG) application to provide real-time responses based on a large document corpus. The application needs to retrieve relevant context from a vector database and serve these features to a Generative AI model for prompt construction. Which of the following resources are necessary to serve features for this RAG application?
- A
A vector database supporting similarity search for embeddings
- B
A feature store to manage historical feature values
- C
A GPU-enabled compute cluster for model inference
- D
A pipeline to periodically update embeddings in the vector database
- E
A distributed file system to store raw training data
Show answer and explanation
Correct answers: A, C, D
Explanation
In a RAG application, serving features involves retrieving relevant context from a vector database and using this context in the Generative AI model's prompt. A vector database is required for efficient similarity search on embeddings, and a GPU-enabled compute cluster is needed for fast inference. Additionally, a pipeline to update embeddings ensures the vector database remains current. A feature store and distributed file system, while useful in other machine learning contexts, do not directly contribute to serving features in this specific scenario.
- A. Correct.
A vector database is critical for storing and retrieving embeddings through similarity search, which is essential for retrieving relevant context in a RAG application.
- B. Incorrect.
A feature store is typically used for managing structured features in machine learning applications, but it is not directly relevant for serving embeddings or real-time retrieval in a RAG application.
- C. Correct.
A GPU-enabled compute cluster is necessary for performing efficient inference with the Generative AI model, which is a core part of the RAG pipeline.
- D. Correct.
A pipeline to periodically update embeddings ensures the vector database remains up-to-date with the latest knowledge, which is crucial for maintaining the accuracy of retrieved context in the RAG application.
- E. Incorrect.
A distributed file system is useful for storing raw training data, but it is not relevant to real-time feature serving in a RAG application.