Databricks Generative AI Engineer Associate Question 2
Select 3You are designing a generative AI application for real-time customer support within a Databricks environment. The application needs to provide accurate and contextually relevant responses while ensuring low latency. Which design considerations should you prioritize to achieve this goal?
- A
Use a lightweight pre-trained model fine-tuned on domain-specific data.
- B
Deploy the generative AI model to a high-latency batch processing pipeline.
- C
Implement a caching mechanism to store frequently asked questions and their responses.
- D
Incorporate prompt engineering to improve the response quality of the model.
- E
Avoid using any optimization techniques to ensure model output remains unaltered.
Show answer and explanation
Correct answers: A, C, D
Explanation
To design a real-time customer support application with generative AI, prioritizing a lightweight fine-tuned model, implementing caching, and leveraging prompt engineering are critical. These strategies collectively ensure accurate, contextually relevant responses while maintaining low latency. Avoiding optimizations or using high-latency architectures would hinder the application's effectiveness in a real-time scenario.
- A. Correct.
Using a lightweight pre-trained model fine-tuned on domain-specific data ensures the application delivers accurate and relevant responses while maintaining low latency. Large models may introduce unnecessary latency and require more computational resources.
- B. Incorrect.
Deploying the generative AI model to a high-latency batch processing pipeline is unsuitable for real-time use cases, as it contradicts the requirement for low latency.
- C. Correct.
Implementing a caching mechanism can significantly reduce response time for frequently asked questions, aligning with the goal of low latency.
- D. Correct.
Incorporating prompt engineering improves the quality of responses by guiding the model more effectively, which is essential for providing contextually relevant answers.
- E. Incorrect.
Avoiding optimization techniques is counterproductive, as optimizations (e.g., model distillation, prompt engineering) are often necessary to balance performance and accuracy in production environments.