Snowflake Cortex provides selected machine-learning and large-language-model capabilities directly through Snowflake interfaces. It lets data teams apply forecasting, anomaly detection, classification, sentiment scoring, prompt completion, summarization, translation, answer extraction, and vector embedding without leaving Snowflake SQL workflows. section 1.4
Use Cortex for pipeline tasks such as:
- Applying time-series forecasting, anomaly detection, or classification to data in Snowflake. section 1.4
- Augmenting records with sentiment scores, summaries, translations, extracted answers, or vector embeddings. section 1.4
- Calling an LLM from procedural logic, transforming its response, and saving the result to a Snowflake table. section 7.5
LLM-augmented pipeline
A pipeline can send a clearly specified prompt to a Cortex LLM function, parse and flatten the JSON response for SQL use, remove unprintable characters when necessary, and persist transformed results in a table. Prompt refinement may require several iterations, and generated results should be reviewed for accuracy. section 7.5
