COF-C03 Question 129
Single answerAI SQL functionsA data engineering team stores customer support tickets in a Snowflake table and wants analysts to classify each ticket into one of several predefined categories such as BILLING, SHIPPING, or TECHNICAL. The team wants a solution that can be called directly from SQL without building and deploying a custom model. Which AI SQL function is the best fit for this requirement?
- A
Use AI_CLASSIFY on the ticket text and provide the allowed category labels
- B
Use AI_COMPLETE to generate a category name from the ticket text with no additional constraints
- C
Use AI_EMBED to convert the ticket text into vectors and treat the vector values as category labels
- D
Use AI_FILTER to remove tickets that do not match known categories
Show answer and explanation
Correct answer: A
Explanation
The best answer is AI_CLASSIFY because the scenario is a classic text classification problem with a known set of valid output labels. Snowflake AI SQL functions are intended to let users apply LLM-powered capabilities directly in SQL, and classification is a distinct use case from generation, filtering, or embeddings. AI_COMPLETE is more appropriate when open-ended text generation is needed, while AI_EMBED supports semantic vector representations for similarity search and related patterns. For SnowPro Core, candidates should recognize the practical fit of each AI SQL function and choose the one that most directly satisfies the business requirement with the least custom implementation.
- A. Correct.
Correct. AI_CLASSIFY is designed for classification tasks in SQL where input text must be assigned to one of a defined set of labels. In this scenario, the analysts already know the allowed categories, so providing those labels to AI_CLASSIFY is the most direct and reliable approach. This matches the requirement to classify unstructured ticket text using a built-in AI SQL function rather than training or deploying a custom model.
- B. Incorrect.
Incorrect. AI_COMPLETE is intended for text generation and general completion-style prompts. Although a prompt could ask for a category name, that approach is less controlled and may return inconsistent output formats, unexpected labels, or extra text. For a bounded classification use case with predefined labels, AI_CLASSIFY is the more appropriate function.
- C. Incorrect.
Incorrect. AI_EMBED generates vector embeddings for semantic similarity and downstream search or ML workflows. Embeddings are not human-readable category labels and do not directly perform classification by themselves. A user might choose this option because embeddings are often used in AI pipelines, but they would still need additional logic or a model to map vectors to categories.
- D. Incorrect.
Incorrect. AI_FILTER is not the right function for assigning one of several business categories to each support ticket. Filtering is about selecting or excluding content based on criteria, not returning a single class label from a predefined list. This option reflects a common confusion between content screening and structured classification.