SnowPro Associate: Platform Question 314
Single answer3.5 Explain how to use Snowflake Cortex LLM functions.A data engineering team stores thousands of customer support case notes in a Snowflake table named SUPPORT_CASES. They want analysts to generate a concise summary of each case directly in SQL without moving data to an external application. The solution should use Snowflake-managed large language model capabilities and return one summary per row. Which SQL approach best meets this requirement?
- A
Use the SNOWFLAKE.CORTEX.SUMMARIZE function in a SELECT statement against the case note column.
- B
Create a materialized view with a CALL statement to an external LLM API for each row in SUPPORT_CASES.
- C
Use COPY INTO to export the case notes to stage files, then apply summarization during file unloading.
- D
Use a standard SQL aggregate such as LISTAGG on the case note column to produce AI-generated summaries.
Show answer and explanation
Correct answer: A
Explanation
The best answer is to use SNOWFLAKE.CORTEX.SUMMARIZE directly in a SELECT statement, for example: SELECT CASE_ID, SNOWFLAKE.CORTEX.SUMMARIZE(CASE_NOTE) AS CASE_SUMMARY FROM SUPPORT_CASES; Snowflake Cortex LLM functions are intended to let users perform tasks such as summarization and text generation within SQL, without exporting data to external AI services. This aligns with Snowflake best practices for minimizing data movement and using managed platform capabilities. On the SnowPro Associate exam, candidates should recognize that Cortex functions are invoked directly in SQL for supported AI tasks, whereas commands like COPY INTO and traditional SQL string functions do not provide LLM behavior.
- A. Correct.
Correct. Snowflake Cortex provides SQL functions for LLM-powered tasks directly inside Snowflake. SNOWFLAKE.CORTEX.SUMMARIZE is designed to generate a concise summary from input text, making it appropriate for summarizing each support case note in a SELECT query. This keeps data in Snowflake and supports row-by-row inference in SQL.
- B. Incorrect.
Incorrect. This approach does not best meet the stated requirement because it moves processing outside the built-in Snowflake Cortex LLM functions and adds unnecessary complexity. Materialized views are also not the intended mechanism for invoking external APIs row-by-row. The scenario specifically asks for Snowflake-managed LLM capabilities directly in SQL.
- C. Incorrect.
Incorrect. COPY INTO is used for loading or unloading data, not for applying Cortex LLM summarization to text. Exporting data to stage files would also violate the goal of performing summarization directly in Snowflake SQL without moving data to an external workflow.
- D. Incorrect.
Incorrect. LISTAGG concatenates strings; it does not perform any semantic summarization or LLM-based processing. A candidate might choose this if they confuse text manipulation with generative AI capabilities, but it cannot create concise meaning-based summaries.