SnowPro Specialty: Gen AI Question 46
Select 21.2 Outline Gen AI capabilities in Snowflake.A retail company stores product descriptions, customer reviews, and support case notes in Snowflake. The analytics team wants to quickly add generative AI features without moving data out of Snowflake. Their first use cases are: summarizing long support cases, extracting sentiment from reviews, and enabling a natural-language assistant that can answer questions grounded in internal product documents. Which Snowflake capabilities should the team use to meet these requirements with the least custom integration effort?
- A
Use Snowflake Cortex AISQL functions for tasks such as summarization and sentiment analysis directly in SQL.
- B
Use Cortex Search as the retrieval layer for grounding responses on internal product documents for the assistant.
- C
Export the data to an external vector database first, because Snowflake does not provide native search or retrieval capabilities for Gen AI workloads.
- D
Train a custom large language model inside a Snowflake virtual warehouse, because Snowflake warehouses are designed to natively train foundation models.
- E
Use Snowflake Document AI to classify and parse structured fields from documents, then use that as the primary feature for conversational question answering over the document corpus.
Show answer and explanation
Correct answers: A, B
Explanation
The best answer is to combine Cortex AISQL functions with Cortex Search. Cortex AISQL functions let teams perform generative AI tasks such as summarization and related text-processing operations directly where the data already resides, reducing ETL and external service integration. Cortex Search supports retrieval over enterprise content for grounded question answering, which is a core pattern for internal assistants and RAG-style applications. Document AI is a separate Snowflake capability focused on understanding and extracting information from documents, not on serving as the retrieval engine for conversational assistants. Likewise, Snowflake does not require exporting data to an external vector database for every Gen AI use case, and Snowflake warehouses are not the mechanism for training foundation models from scratch. These distinctions align with Snowflake documentation and product positioning for Cortex AI, Cortex Search, and Document AI.
- A. Correct.
Correct. Snowflake Cortex provides AI functions accessible from SQL for common generative and language tasks, including summarization and sentiment-related text analysis use cases. This is the most direct way to add these capabilities while keeping data in Snowflake and minimizing custom application logic.
- B. Correct.
Correct. Cortex Search is designed to support low-latency retrieval over enterprise data in Snowflake, making it appropriate for retrieval-augmented generation scenarios where an assistant must answer questions grounded in internal documents. It reduces the need to build and manage a separate retrieval stack.
- C. Incorrect.
Incorrect. This reflects a common misconception that Gen AI retrieval always requires a separate external vector database. Snowflake provides native capabilities for AI-powered search and retrieval through Cortex Search, so exporting data first would add unnecessary complexity and data movement.
- D. Incorrect.
Incorrect. Snowflake virtual warehouses are compute resources for executing workloads such as SQL, data engineering, and ML-related operations, but they are not positioned as a native environment for training custom foundation models. Snowflake's Gen AI capabilities focus on consuming and operationalizing AI functions and services rather than using warehouses to train LLMs from scratch.
- E. Incorrect.
Incorrect. Document AI is useful for extracting information from documents, especially when dealing with document processing workflows. However, it is not the primary capability for building a conversational assistant that answers grounded questions across a document corpus. For that retrieval-based assistant pattern, Cortex Search is the more appropriate Snowflake capability.