Databricks Generative AI Engineer Associate Question 5
Select 3You are designing a generative AI application on Databricks that generates personalized product recommendations for users. The application needs to process user behavior data in real-time, utilize a fine-tuned large language model (LLM), and serve recommendations with low latency. Which components should you include in your design to meet these requirements?
- A
A Databricks Delta Live Tables pipeline to process and prepare real-time user behavior data
- B
A fine-tuned LLM hosted in a Databricks Model Serving endpoint
- C
A batch processing job to periodically update recommendations
- D
A distributed feature store to manage and serve user features for the model
- E
A Databricks SQL endpoint to serve recommendations directly to users
Show answer and explanation
Correct answers: A, B, D
Explanation
To design a low-latency generative AI application for personalized recommendations, you need to process user behavior data in real-time using Delta Live Tables, host the fine-tuned LLM on a Databricks Model Serving endpoint for efficient inference, and use a distributed feature store to manage and serve user-specific features. Batch processing and SQL endpoints are not suitable for the real-time and low-latency requirements specified in this scenario.
- A. Correct.
This is correct. Databricks Delta Live Tables can process real-time data streams, making it suitable for handling user behavior data in real-time.
- B. Correct.
This is correct. Hosting a fine-tuned LLM on a Databricks Model Serving endpoint allows for low-latency inference, which is critical for serving recommendations in real-time.
- C. Incorrect.
This is incorrect. While batch jobs can update recommendations, they do not meet the real-time processing requirement outlined in the scenario.
- D. Correct.
This is correct. A distributed feature store enables real-time management and retrieval of user features, which is necessary for generating personalized recommendations efficiently.
- E. Incorrect.
This is incorrect. While Databricks SQL is useful for querying structured data, it is not optimized for real-time recommendations or LLM inference.