COF-C03 Question 126
Single answerSnowflake CortexA product analytics team stores customer support case text in a Snowflake table and wants to add sentiment labels directly in SQL so analysts can include the results in dashboards. The team does not want to build or manage a separate ML pipeline outside Snowflake. Which approach should the team use to meet this requirement with the least operational overhead?
- A
Use a Snowflake Cortex AI function such as SENTIMENT directly in a SQL query against the support case text column.
- B
Export the support case table to an external Python environment, train a custom sentiment model, and write the results back to Snowflake.
- C
Create a materialized view that automatically trains a sentiment model whenever new rows are inserted.
- D
Use Snowpipe Streaming to convert unstructured text into sentiment labels during ingestion without calling any SQL functions.
Show answer and explanation
Correct answer: A
Explanation
The best answer is to use a Snowflake Cortex AI function directly in SQL. For SnowPro Core, candidates should recognize that Snowflake Cortex enables built-in AI capabilities within Snowflake, allowing users to perform tasks like text analysis without creating external machine learning infrastructure. This aligns with the scenario's goals: keep the workflow in Snowflake, enable analysts to use SQL, and minimize operational overhead. By contrast, exporting data to external ML systems increases complexity, materialized views do not provide model training, and Snowpipe Streaming is an ingestion technology rather than an AI inference feature. Snowflake documentation describes Cortex as providing SQL-accessible AI functions for common generative AI and language tasks, making it a practical fit for this type of use case.
- A. Correct.
Correct. Snowflake Cortex provides AI functions that can be invoked directly from SQL to analyze text, including sentiment-related use cases. This is the lowest-overhead option because it keeps processing in Snowflake and does not require the team to build, train, deploy, or operate a separate ML pipeline.
- B. Incorrect.
Incorrect. While exporting data and using Python could produce sentiment labels, it adds significant operational overhead and contradicts the requirement to avoid a separate ML pipeline outside Snowflake. This is a common approach in traditional architectures, but Snowflake Cortex is designed to reduce that complexity for supported AI tasks.
- C. Incorrect.
Incorrect. Materialized views do not train machine learning models. They store precomputed query results for performance optimization. This option reflects a misconception that Snowflake performance features also provide automated model training capabilities.
- D. Incorrect.
Incorrect. Snowpipe Streaming is used for low-latency data ingestion into Snowflake, not for performing sentiment analysis by itself. It can help bring text data into Snowflake quickly, but sentiment labeling would still require a separate step, such as calling a Cortex AI function in SQL.