SnowPro Specialty: Gen AI Question 180
Single answer2.4 Use Snowflake Cortex functions in data pipelines.A retail company stores customer support tickets in a Snowflake table and wants to enrich each new ticket with a short summary and sentiment label as part of a daily SQL-based pipeline. The data engineering team wants to keep processing inside Snowflake and avoid moving data to external services. Which approach best meets these requirements?
- A
Create a SQL transformation that uses Snowflake Cortex functions such as SUMMARIZE and SENTIMENT in an INSERT ... SELECT or MERGE step to populate enriched columns in a downstream table.
- B
Export new tickets to an external Python service, call a third-party LLM API for summarization and sentiment, and load the results back into Snowflake because Cortex functions cannot be used in SQL pipelines.
- C
Create a materialized view that directly stores the output of Cortex functions for each incoming row, since LLM function outputs are the recommended pattern for persisted pipeline enrichment.
- D
Use Cortex Search to generate summaries and sentiment labels during ETL, because search services are designed for text generation inside batch pipelines.
Show answer and explanation
Correct answer: A
Explanation
The best answer is to use Snowflake Cortex functions directly in SQL transformation steps within the pipeline. Snowflake Cortex AISQL functions are designed so teams can perform tasks such as summarization, sentiment analysis, classification, translation, and other LLM-powered enrichments without exporting data from Snowflake. In a production pipeline, this is commonly done with INSERT ... SELECT, MERGE, tasks, dynamic tables, or other SQL orchestration patterns, depending on the architecture. This approach minimizes data movement, simplifies governance, and keeps processing under Snowflake security controls. By contrast, external API workflows add complexity and violate the scenario requirement, materialized views are not the standard persistence pattern for this type of enrichment, and Cortex Search serves retrieval use cases rather than text generation or sentiment labeling. Candidates should recognize that Cortex functions can be embedded directly in SQL-based data pipelines for practical in-platform AI enrichment.
- A. Correct.
Correct. Snowflake Cortex provides SQL functions that can be called directly in data pipelines to enrich data in place. For this scenario, a SQL statement such as INSERT ... SELECT, CREATE TABLE AS SELECT, or MERGE can apply functions like SUMMARIZE and SENTIMENT to ticket text and write the results into target columns. This keeps processing inside Snowflake and aligns with the requirement to avoid external movement of data.
- B. Incorrect.
Incorrect. Although exporting data to an external application is technically possible, it does not meet the stated goal of keeping processing inside Snowflake. The key misconception is that Cortex functions cannot be used in SQL pipelines. In reality, Cortex AISQL functions are specifically intended to be invoked from SQL for in-database AI enrichment workflows.
- C. Incorrect.
Incorrect. Materialized views are not the recommended mechanism for persisting LLM-generated enrichment results in this type of pipeline. The misconception is treating Cortex function outputs like deterministic expressions suited for materialized view storage and refresh semantics. In practice, batch enrichment is better implemented with explicit pipeline steps that write results to tables.
- D. Incorrect.
Incorrect. Cortex Search is for building low-latency search and retrieval experiences over Snowflake data, not for generating summaries or sentiment labels during ETL. The misconception is confusing retrieval/search capabilities with generative and analytical Cortex SQL functions.