SnowPro Specialty: Gen AI Question 166
Single answer2.3 Build chat interfaces to interact with data in Snowflake.A retail analytics team wants to give business users a chat interface that can answer questions such as "Which product categories had the highest return rate last quarter?" using governed data already stored in Snowflake. The team wants the solution to minimize custom orchestration code, respect Snowflake role-based access controls on underlying tables, and allow the assistant to generate SQL when appropriate. Which approach is the BEST fit?
- A
Create a Cortex Analyst semantic model over the relevant business data and expose it through a chat application so natural-language questions can be translated into governed SQL against Snowflake data.
- B
Build a Cortex Search Service on the transaction tables and use it as the primary mechanism for computing return-rate metrics directly from natural-language prompts.
- C
Fine-tune a custom LLM outside Snowflake on exported retail data and let the model answer directly from training weights, since this avoids querying governed tables at runtime.
- D
Use Snowflake Notebooks as the end-user chat interface because notebooks automatically convert business questions into production-ready SQL for nontechnical users.
Show answer and explanation
Correct answer: A
Explanation
The best answer is to use Cortex Analyst with a semantic model. For chat interfaces that interact with structured data in Snowflake, Cortex Analyst is the purpose-built capability for turning natural-language questions into SQL grounded in Snowflake data and business semantics. This reduces the amount of custom orchestration needed compared with building an end-to-end NL-to-SQL solution manually. It also fits enterprise governance requirements because queries execute against Snowflake objects under established permissions and policies. By contrast, Cortex Search is better suited for retrieval over documents or text rather than metric computation from structured relational data. External fine-tuning on exported data weakens governance and freshness, and Snowflake Notebooks are a development environment rather than the best-fit governed chat interface for business users. Candidates should recognize the distinction between conversational analytics over structured data, where Cortex Analyst is appropriate, and retrieval-based chat over unstructured content, where Cortex Search is more appropriate.
- A. Correct.
Correct. Cortex Analyst is designed for conversational analytics over structured data in Snowflake. By defining a semantic model, teams can map business concepts, metrics, and relationships so natural-language questions are translated into SQL against governed Snowflake data. This aligns with the requirement to minimize custom orchestration while respecting existing access controls on the underlying objects.
- B. Incorrect.
Incorrect. Cortex Search Service is intended for low-latency retrieval over indexed text and semi-structured content, making it useful for RAG-style retrieval experiences. It is not the primary tool for generating governed analytical SQL over structured business metrics such as return rate calculations from relational tables. Someone might choose this because it supports conversational experiences, but it solves a different problem.
- C. Incorrect.
Incorrect. Exporting data and relying on a fine-tuned external model conflicts with the requirement to respect Snowflake governance at query time. A model answering from training weights can become stale, may not reflect current data, and does not inherently enforce Snowflake RBAC on live table access. This is a common misconception when teams confuse model training with governed analytical querying.
- D. Incorrect.
Incorrect. Snowflake Notebooks are useful for development, exploration, and collaboration, but they are not a purpose-built end-user chat interface that automatically turns business-user questions into governed SQL. A notebook could be part of prototyping, but it does not satisfy the requirement as directly or appropriately as Cortex Analyst.