Databricks Generative AI Engineer Associate Question 1
Select 3You are designing a generative AI application on Databricks that uses a pre-trained large language model (LLM) to provide text summarization services for customer support tickets. The application needs to ensure scalability, low latency, and compliance with enterprise data governance policies. Which design considerations should you prioritize while building this application on Databricks?
- A
Implement model serving with Databricks Model Serving to deploy the LLM for low-latency predictions.
- B
Use Unity Catalog to ensure proper data lineage and access control for customer support ticket data.
- C
Store customer support tickets in local file systems for faster processing.
- D
Leverage Delta Lake for efficient storage and management of support ticket data with ACID transactions.
- E
Avoid fine-tuning the pre-trained LLM and use it as-is to minimize model complexity.
Show answer and explanation
Correct answers: A, B, D
Explanation
To design a scalable and efficient generative AI application on Databricks, it is critical to prioritize model serving for low latency, use Unity Catalog for data governance, and leverage Delta Lake for efficient data management. These components ensure the application meets enterprise requirements for scalability, compliance, and performance.
- A. Correct.
Correct: Databricks Model Serving provides a scalable and low-latency way to deploy and serve LLMs, which is crucial for real-time applications like text summarization.
- B. Correct.
Correct: Unity Catalog ensures compliance with enterprise data governance policies by providing centralized data access control and lineage tracking.
- C. Incorrect.
Incorrect: Storing data in local file systems does not align with scalability and enterprise data governance best practices. Databricks recommends cloud-based storage solutions like Delta Lake.
- D. Correct.
Correct: Delta Lake enables efficient handling of large-scale data with ACID transactions, ensuring data integrity and scalability for analytics and AI workflows.
- E. Incorrect.
Incorrect: While using the pre-trained LLM as-is may simplify the design, it is not a priority for this scenario. Fine-tuning may still be required to improve performance on domain-specific customer support data.