Databricks Generative AI Engineer Associate Question 237
Select 3You are tasked with building a Retrieval-Augmented Generation (RAG) application that uses a large language model (LLM) to provide accurate, real-time answers based on your company's proprietary data. To serve features effectively for this application, which resources should you prioritize for optimal performance and scalability?
- A
A vector database to store and query embeddings efficiently.
- B
A feature store to manage and serve real-time features.
- C
A caching layer to reduce latency for frequently accessed features.
- D
A distributed compute cluster for training the LLM.
- E
A serverless function to orchestrate model inference and retrieval.
Show answer and explanation
Correct answers: A, B, C
Explanation
Serving features efficiently in a Retrieval-Augmented Generation (RAG) pipeline requires resources that support fast and scalable access to embeddings, real-time features, and frequently accessed data. A vector database, feature store, and caching layer are critical components for achieving these goals. While training infrastructure and orchestration tools are important in a broader AI pipeline, they are not directly tied to feature serving in this context.
- A. Correct.
A vector database is essential for storing and retrieving embeddings generated during the retrieval phase of a RAG pipeline.
- B. Correct.
A feature store helps in managing and serving real-time features, which are critical for ensuring that the RAG application has access to accurate and up-to-date data.
- C. Correct.
A caching layer can significantly reduce latency by providing quick access to frequently used features, thereby improving the performance of the application.
- D. Incorrect.
A distributed compute cluster is more relevant to training an LLM rather than serving features in a RAG pipeline. This is not directly related to the feature-serving aspect of the application.
- E. Incorrect.
While a serverless function may help with orchestration, it is not a primary resource for serving features in a RAG application.