COF-C03 Question 127
Single answerSnowflake CortexA support analytics team stores customer case notes in a Snowflake table and wants to build a quick proof of concept that labels each note with a short sentiment summary directly in SQL. The team does not want to provision or manage ML infrastructure, and they want the processing to remain inside Snowflake. Which approach best meets these requirements?
- A
Use a Snowflake Cortex AISQL function such as SENTIMENT or SUMMARIZE directly in a SQL query against the case notes table.
- B
Export the case notes to an external machine learning platform, generate labels there, and load the results back into Snowflake.
- C
Create a materialized view with a Python UDF that downloads a public language model from the internet at query time.
- D
Use Snowpipe Streaming to automatically infer sentiment labels as data is ingested, because Cortex functions only work during data loading.
Show answer and explanation
Correct answer: A
Explanation
Snowflake Cortex is designed to bring managed AI capabilities directly into Snowflake so users can call language functions from SQL without provisioning or managing separate machine learning infrastructure. For a proof of concept that needs to analyze text already stored in Snowflake, AISQL functions such as SENTIMENT and SUMMARIZE are the most direct and operationally simple approach. This aligns with Snowflake best practices of minimizing data movement and using built-in platform capabilities where possible. External ML platforms can work, but they add complexity and do not satisfy the requirement as well. Snowpipe Streaming addresses ingestion, not in-database text inference. Refer to Snowflake documentation for Cortex AISQL functions and their use directly in SQL queries against Snowflake tables.
- A. Correct.
Correct. Snowflake Cortex provides AISQL functions that can be invoked directly from SQL to perform tasks such as sentiment analysis and summarization on data stored in Snowflake. This fits the scenario because the team can enrich text data without provisioning separate ML infrastructure and can keep processing inside Snowflake.
- B. Incorrect.
Incorrect. While this could produce sentiment labels, it does not meet the requirement to avoid managing external ML infrastructure and keep processing inside Snowflake. This option adds data movement, operational overhead, and governance complexity compared with using Cortex directly in Snowflake.
- C. Incorrect.
Incorrect. This is not the best fit and includes an invalid assumption. A Python UDF is not the recommended way to quickly apply Snowflake Cortex language capabilities, and downloading arbitrary public models from the internet at query time is not how Snowflake Cortex is used. The scenario specifically asks for a managed in-platform capability.
- D. Incorrect.
Incorrect. Snowpipe Streaming is for ingesting streaming data into Snowflake, not for automatically performing language inference because Cortex functions are limited to load time. Cortex AISQL functions are invoked in SQL queries and are not restricted to ingestion workflows.