SnowPro Specialty: Gen AI Question 183
Single answerSQL interfaceA data engineering team is building a SQL-only workflow in Snowflake to classify incoming customer support tickets and generate draft responses for agents to review. The team wants to stay entirely within SQL so analysts can prototype quickly in worksheets, and they need the model output to be available directly in query results for downstream joins and reporting. Which approach best meets this requirement?
- A
Use Snowflake Cortex AISQL functions in a SELECT statement to send ticket text to a supported model and return the generated classification and draft response as SQL query results.
- B
Create a Python stored procedure because Snowflake SQL cannot invoke large language models directly; only procedural code can access generative AI models.
- C
Export ticket text from Snowflake to an external application, call the model there, and reload the results because SQL interfaces in Snowflake support only embeddings, not text generation.
- D
Use a standard SQL UDF with no external service integration to generate the draft response locally inside the warehouse compute layer.
Show answer and explanation
Correct answer: A
Explanation
The best answer is to use Snowflake Cortex AISQL functions directly in SQL. This is specifically designed for SQL-first AI workflows where users want to invoke supported models from worksheets, queries, and pipelines without leaving Snowflake or writing application code. In a real-world scenario like support-ticket classification and draft-response generation, this approach minimizes data movement, simplifies governance, and keeps outputs in relational form for downstream analytics. Snowflake documentation for Cortex AI and AISQL describes using SQL functions to perform LLM-powered tasks such as generation, classification, summarization, and related inference directly in SQL. The key exam takeaway is that the SQL interface is not limited to traditional analytics; it can also be the primary interface for GenAI operations within Snowflake when using the appropriate built-in AI functions.
- A. Correct.
Correct. Snowflake provides Cortex AISQL capabilities that let users invoke supported AI models directly from SQL. This is the most appropriate approach when the goal is to remain in a SQL-only workflow and have outputs returned directly in result sets for further SQL processing such as joins, filtering, and reporting. This matches the scenario's requirement for analyst-friendly prototyping in worksheets and downstream SQL consumption.
- B. Incorrect.
Incorrect. This reflects a common misconception that generative AI access in Snowflake requires Python or procedural logic. While Python stored procedures can be used for many advanced workflows, Snowflake also exposes AI functionality directly through SQL functions, so procedural code is not required for this use case.
- C. Incorrect.
Incorrect. This option adds unnecessary data movement and operational complexity. It also incorrectly assumes that Snowflake SQL supports only embeddings and not text generation. In practice, Snowflake's SQL interface can be used for generative AI tasks through Cortex AISQL functions, making external export/reload unnecessary for the stated requirement.
- D. Incorrect.
Incorrect. A standard SQL UDF does not independently host or run a large language model inside warehouse compute with no external model integration. This distractor plays on the misconception that SQL UDFs can natively perform LLM inference without calling a supported AI service. In Snowflake, generative AI inference is provided through specific AI features and functions rather than ordinary local SQL UDF execution alone.