Databricks Generative AI Engineer Associate Question 235
Select 3You are designing a Retrieval-Augmented Generation (RAG) application on Databricks. The application requires serving features from a feature store that can provide real-time embeddings for a large-scale question-answering system. Which resources are essential for serving features efficiently in this scenario?
- A
A low-latency online feature store
- B
A batch processing engine for periodic feature updates
- C
A vector database for storing and querying embeddings
- D
A scalable REST API to serve features to the application
- E
A model training cluster for fine-tuning embeddings
Show answer and explanation
Correct answers: A, C, D
Explanation
In a RAG application, serving real-time features efficiently requires a low-latency online feature store for immediate access, a vector database for embedding storage and retrieval, and a scalable REST API for feature delivery. Batch processing and model training are not critical to the real-time serving process but may be part of other stages in the pipeline.
- A. Correct.
A low-latency online feature store is critical for serving real-time features to the RAG application efficiently. Without this, the application may experience delays in feature retrieval.
- B. Incorrect.
While batch processing is useful for periodic updates, it is not essential for serving real-time features in the context of the RAG application.
- C. Correct.
A vector database is necessary to store and retrieve embeddings used in the RAG application, especially when working with similarity searches or dense vector queries.
- D. Correct.
A scalable REST API is required to serve features to external applications, ensuring smooth integration and real-time delivery.
- E. Incorrect.
A model training cluster is not directly involved in serving features. It is typically used during the model development phase, not during feature serving.