Snowflake Cortex AI Features
Free SnowPro Specialty: Gen AI practice — 6 questions on Snowflake Cortex AI Features, with explanations. No sign-up. Full 12-question mixed test →
A financial services firm stores 2 million scanned PDF contracts in an internal stage. They want employees to ask natural-language questions like 'What is the termination clause in the Acme Corp contract?' and receive exact excerpted text with source citations. Which Cortex-based architecture BEST meets this requirement?
This is a classic unstructured RAG pipeline: parse the PDFs into text, chunk appropriately, embed the chunks, and index them with Cortex Search so users get semantically relevant excerpts with citations back to source documents.
A data team needs to add a numeric column to a Snowflake table that scores each customer support ticket's overall sentiment from -1 to 1 for a downstream analytics dashboard. Which single Cortex LLM function should they call directly in a SQL SELECT statement to produce this score?
SENTIMENT returns a single continuous float score between -1 and 1 representing overall sentiment for the input text, which is exactly the numeric output the dashboard needs.
A developer creates a Cortex Search service with TARGET_LAG = '1 hour' over a base view built from a change-tracking-enabled table. After loading 50,000 new rows into the base table, users report the search results are missing the new documents even 30 minutes after the load. What is the MOST likely explanation?
TARGET_LAG behaves like it does for dynamic tables — it defines the maximum acceptable staleness/refresh interval for the underlying index, so a delay of up to the configured lag after a base table load is expected, not an error.
A team wants a Cortex LLM to consistently produce legal documents in a highly specific in-house format and tone that has proven difficult to achieve through prompt engineering alone, even with detailed instructions and few-shot examples. They have 5,000 labeled example documents available. Which Cortex capability is the MOST appropriate next step?
Fine-tuning is designed for exactly this scenario: prompt engineering has plateaued, a well-defined output style/format is needed consistently, and sufficient labeled training data exists to teach the model the desired behavior.
A retail analytics team wants business users to ask questions like 'What were total sales by region last quarter?' directly against Snowflake tables using natural language, with Cortex Analyst generating and executing the correct SQL. During setup, the team defines table relationships, column descriptions, synonyms, and sample verified queries. Where must this configuration be stored for Cortex Analyst to use it?
Cortex Analyst requires a semantic model defined as a YAML file, stored on a Snowflake stage, that describes tables, columns, relationships, synonyms, and verified queries so it can reliably translate natural language into correct SQL.
A team wants to fine-tune a Cortex LLM for a specialized medical coding task using SNOWFLAKE.CORTEX.FINETUNE with a chosen base model parameter. Which of the following is a valid consideration they must account for BEFORE starting the job?
Cortex Fine-Tuning is a serverless, credit-metered operation where cost scales with training data volume and the chosen base model, and only a documented subset of text-generation base models is eligible for fine-tuning.
Ready for the real thing?
The full course has two full-length practice tests, video lessons for every exam domain, hands-on labs and detailed answer explanations.
Start my full course on Udemy →