ARA-C01 Question 196
Single answerSnowflake ML functionsA retail company stores customer support tickets in Snowflake and wants to add sentiment scores and short summaries directly in SQL so analysts can use the results in downstream dashboards. The architects want to minimize data movement, avoid managing external ML infrastructure, and keep the solution easy to govern with Snowflake roles. Which approach best meets these requirements?
- A
Use Snowflake Cortex AISQL functions such as SENTIMENT and SUMMARIZE directly against the ticket text columns, and persist the results in Snowflake tables for downstream reporting.
- B
Export the ticket data to an external Python service, run an open-source sentiment model and summarizer there, and write the predictions back to Snowflake because Snowflake ML functions require external compute.
- C
Create a Java UDF in Snowflake to call a third-party LLM endpoint for both sentiment and summarization, because Snowflake does not provide SQL functions for these tasks.
- D
Train a custom forecasting model with Snowflake ML and use it to infer both sentiment and summaries, because forecasting models can generalize to any text analytics requirement.
Show answer and explanation
Correct answer: A
Explanation
The best answer is to use Snowflake Cortex AISQL functions directly in Snowflake. For this scenario, the key architectural drivers are minimizing data movement, simplifying operations, and keeping governance within Snowflake. Snowflake provides built-in SQL functions for AI use cases such as sentiment analysis and summarization, which are better aligned than exporting data to external services or misusing other ML model types. From an architect perspective, this design reduces integration points, keeps sensitive text in-platform, and allows standard Snowflake access controls and auditing to apply. This is consistent with Snowflake best practices of processing data as close to where it resides as possible and using native platform capabilities when they satisfy the requirement.
- A. Correct.
Correct. Snowflake provides Cortex AISQL functions for common AI/ML tasks directly in SQL, including sentiment analysis and text summarization. This approach keeps the data in Snowflake, reduces operational complexity, and aligns with the requirement to avoid building and maintaining separate ML infrastructure. It also supports centralized governance using Snowflake security and role-based access controls.
- B. Incorrect.
Incorrect. While exporting data to an external service is technically possible, it does not meet the stated goal of minimizing data movement and avoiding external ML infrastructure. It also adds integration, security, latency, and operational overhead. The misconception is assuming Snowflake cannot perform these text AI tasks natively in SQL.
- C. Incorrect.
Incorrect. A Java UDF or external integration would add unnecessary complexity for this use case. Snowflake already offers built-in SQL-accessible AI functions for sentiment and summarization, so building and governing a custom callout path is not the best architectural choice here. This distractor reflects a common assumption that LLM-based features must be manually integrated.
- D. Incorrect.
Incorrect. Forecasting models are intended for time-series prediction, not for natural language tasks such as sentiment classification or summarization. This option confuses different categories of ML capabilities. Using a forecasting model for text analytics would not be an appropriate or supported design.