SnowPro Specialty: Gen AI Question 199
Single answerEnvironment setupA data engineering team is preparing a new Snowflake environment so application developers can build a Retrieval-Augmented Generation (RAG) prototype using Snowflake-managed LLM features. The security team requires least-privilege access and wants to avoid unnecessary external infrastructure. The developers need to parse internal documents, generate embeddings, and run SQL/Python workloads inside Snowflake. Which setup step is the MOST appropriate to complete first to enable this work?
- A
Create a warehouse, database, schema, and grant developers the required role privileges to use Cortex functions and build in-container SQL/Python workloads in Snowflake
- B
Provision a separate GPU cluster outside Snowflake and configure network peering before any Snowflake objects are created
- C
Create an external function integration to a third-party LLM service, because Snowflake-managed LLM features require an external API endpoint
- D
Enable cross-cloud replication for the account, because document parsing and embedding functions are only available after replication is configured
Show answer and explanation
Correct answer: A
Explanation
The best answer is to begin with core Snowflake environment setup: create the required database objects and compute resources, then apply role-based access control so developers can use the necessary Snowflake capabilities with least privilege. In a Snowflake-native GenAI workflow, this typically means enabling developers to run SQL and Python in Snowflake and granting access to relevant Cortex functions or features that support tasks such as document processing and embedding generation. This approach avoids unnecessary external infrastructure and matches Snowflake operational best practices: set up the account environment, warehouses, databases/schemas, and privileges first, then build the RAG pipeline. Snowflake documentation for Cortex and general security guidance emphasizes RBAC, least privilege, and using Snowflake-managed services where possible to simplify deployment.
- A. Correct.
Correct. For environment setup, the first practical step is to establish the Snowflake objects and access model developers need: compute (warehouse), logical containers (database/schema), and least-privilege role grants. For GenAI work in Snowflake, teams typically need permissions to run SQL and Python workloads and to use Cortex capabilities such as document processing and embedding-related functions where enabled. This aligns with Snowflake best practices of starting with account/database/schema/warehouse setup and role-based access control before building the application workflow.
- B. Incorrect.
Incorrect. Snowflake-managed GenAI features such as Cortex are designed to reduce the need for customers to provision and manage separate GPU infrastructure. A team may choose external infrastructure for other architectures, but it is not the most appropriate first step for a Snowflake-native RAG prototype using managed capabilities.
- C. Incorrect.
Incorrect. This reflects a common misconception that all LLM access from Snowflake must go through external functions. Snowflake Cortex provides managed AI functions directly in Snowflake, so an external API endpoint is not required just to get started with document parsing, embeddings, or prompt-based inference using Snowflake-managed features.
- D. Incorrect.
Incorrect. Cross-cloud replication is unrelated to the initial enablement of document parsing, embedding generation, or SQL/Python development for a RAG prototype. Replication supports business continuity and data distribution scenarios, not basic GenAI environment setup.