Snowflake SnowPro Specialty Gen AI Quick Facts (2025)

Exam Domain Breakdown

Domain 1: Snowflake for Gen AI Overview (26% of the exam)

Define Snowflake’s Gen AI principles, features, and best practices.

• Snowflake Cortex — LLMs
• Snowflake Cortex — Cortex Search
• Snowflake Cortex — Cortex Analyst
• Snowflake Cortex — Cortex Fine-tuning
• Snowflake Cortex — Cortex Agents (Public Preview)
• Snowflake Copilot
• Security, privacy, access, and control principles — Role-Based Access Control (RBAC)
• Security, privacy, access, and control principles — Guardrails
• Security, privacy, access, and control principles — Required privileges
• Security, privacy, access, and control principles — Cortex LLM Functions — Control model access
• Security, privacy, access, and control principles — CORTEX_MODELS_ALLOWLIST parameter
• Different interfaces — Cortex LLM Playground (Public Preview)
• Different interfaces — SQL
• Different interfaces — REST API
• Different ways of bringing your own models into Snowflake (for example, from Hugging Face) — Using Snowflake Model Registry (custom model)
• Different ways of bringing your own models into Snowflake (for example, from Hugging Face) — Using Snowpark Container Services

Section summary: This section builds a strong foundation in core Gen AI capabilities within Snowflake, showcasing services such as Cortex LLMs, Cortex Agents, and the Cortex Copilot. You will learn how interfaces including SQL, REST API, and the Cortex LLM Playground provide flexible ways for different stakeholders to use these capabilities inside their workflows.

Alongside functional knowledge, the focus extends to critical principles of security, privacy, and governance that ensure your solutions are responsible, scalable, and team-ready. This includes RBAC, guardrails, and privileges designed to protect data while still enabling powerful model-driven insights.

Outline Gen AI capabilities in Snowflake.

• Cortex LLM functions (for example, task-specific, general) — Vector-embedding
• Cortex LLM functions (for example, task-specific, general) — Fine-tuning
• Cortex Search — RAG use cases
• Cortex Search — Unstructured data use cases
• Cortex Search — REST APIs
• Cortex Analyst — Semantic model generation — Stored in YAML files in a stage
• Cortex Analyst — Semantic model generation — Stored natively in semantic views (Public Preview)
• Cortex Analyst — Structured/text-to-SQL use cases
• Cortex Analyst — REST APIs
• Cortex Agents (Public Preview) — REST APIs
• Cross-region inference — CORTEX_ENABLED_CROSS_REGION parameter
• Cross-region inference — Considerations (for example, latency, availability)

Section summary: Here, you will explore how multiple Cortex features unlock generative AI across data analysis and application development. By diving into task-specific functions like fine-tuning and vector embeddings, as well as tools like Cortex Analyst and Cortex Search, you will gain insight into building Retrieval Augmented Generation (RAG) use cases and translating text into actionable queries.

Additionally, this section highlights technical considerations such as cross-region inference, APIs, and integration details, which help ensure that solutions are accurate, performant, and ready for enterprise scale. It equips you to connect model functionality with real-world Snowflake workloads.

Domain 2: Snowflake Gen AI & LLM Functions (40% of the exam)

Apply Gen AI and LLM functions in Snowflake.

• Snowflake Cortex — General — COMPLETE
• Snowflake Cortex — General — COMPLETE Structured Outputs
• Snowflake Cortex — Task-specific functions — CLASSIFY_TEXT
• Snowflake Cortex — Task-specific functions — EXTRACT_ANSWER
• Snowflake Cortex — Task-specific functions — PARSE_DOCUMENT
• Snowflake Cortex — Task-specific functions — SENTIMENT
• Snowflake Cortex — Task-specific functions — SUMMARIZE
• Snowflake Cortex — Task-specific functions — TRANSLATE
• Snowflake Cortex — Task-specific functions — EMBED_TEXT_768
• Snowflake Cortex — Task-specific functions — EMBED_TEXT_1024
• Cortex Search
• Cortex Analyst
• Cortex Fine-tuning
• Cortex Agents (Public Preview)
• Vector functions — VECTOR_INNER_PRODUCT
• Vector functions — VECTOR_L1_DISTANCE
• Vector functions — VECTOR_L2_DISTANCE
• Vector functions — VECTOR_COSINE_SIMILARITY
• Helper functions — COUNT_TOKENS
• Helper functions — TRY_COMPLETE
• Helper functions — SPLIT_TEXT_RECURSIVE_CHARACTER
• Choosing a model — Considerations (e.g. capability, latency, and cost)

Section summary: This section dives into the wide variety of Snowflake Cortex functions that help implement generative AI solutions. From general-purpose text completion to task-specific utilities such as extracting answers, summarization, classification, and translation, you will explore capabilities that enhance text intelligence across diverse use cases.

You will also learn how vector functions and helper functions expand your toolkit for embedding data, comparing outputs, and monitoring usage. This practical knowledge ensures that when you choose between models or text functions, your choices align with latency, cost, and capability requirements.

Perform data analysis given a use case.

• Use fully-managed LLMs, RAG, and text-to-SQL services — Unstructured data — CORTEX PARSE_DOCUMENT
• Use fully-managed LLMs, RAG, and text-to-SQL services — Structured data
• Use fully-managed LLMs, RAG, and text-to-SQL services — Cortex Analyst — Cortex Analyst Verified Query Repository (VQR)
• Use fully-managed LLMs, RAG, and text-to-SQL services — Cortex Analyst — Integration with Cortex Search
• Use fully-managed LLMs, RAG, and text-to-SQL services — Cortex Analyst — Suggested Questions
• Use fully-managed LLMs, RAG, and text-to-SQL services — Cortex Analyst — Custom_instructions field
• Performance considerations — Latency (for example, fine-tuning, model size)

Section summary: This part of the exam validates your ability to apply generative AI for real-world data analysis. With tools like Cortex Analyst and Verified Query Repositories, you will see how to streamline workflows and expand the reach of text-to-SQL solutions, including integration with Cortex Search for contextual intelligence.

Performance considerations such as latency and model size are also covered to ensure that your analysis stays efficient and reliable. The combination of unstructured and structured data use cases demonstrates how Snowflake’s LLM-powered services can transform data into insights quickly and effectively.

Build chat interfaces to interact with data in Snowflake.

• Set up the Snowflake environment — Required privileges
• Invoke Cortex functions within the application code (for example, Streamlit)
• Chat conversations — Multi-turn architecture
• Chat conversations — Update parameters

Section summary: This section focuses on creating practical, interactive chat interfaces. You will learn the essential permissions and environment setup steps to run conversational applications powered by Snowflake Cortex, giving end users intuitive access to dataset-driven insights.

Key concepts include multi-turn chat architectures and parameter updates, enabling fluid and intelligent conversations over business data. By leveraging platforms like Streamlit for integration, these skills translate directly into enriched user experiences and real-time AI responses.

Use Snowflake Cortex functions in data pipelines.

• Snowflake Cortex — SQL interface
• Snowflake Cortex — Extracting data from text using COMPLETE — Transcripts
• Snowflake Cortex — Data enrichment
• Snowflake Cortex — Data augmentation
• Snowflake Cortex — Data transformations

Section summary: This section demonstrates how to integrate AI into scalable ETL and ELT pipelines. You will learn how text completion, enrichment, augmentation, and transformation can be automated inside data pipelines, ensuring these operations work seamlessly within your enterprise workflows.

As a result, you gain the skills to elevate pipelines by embedding intelligence directly at the data flow level. This creates powerful opportunities for enhancing both structured and unstructured data in motion.

Run third-party models in Snowflake.

• Using Snowpark Container Services — Environment setup
• Using Snowpark Container Services — Docker images
• Using Snowpark Container Services — Specification files
• Using Snowpark Container Services — Create compute pool
• Using Snowpark Container Services — Create image repository
• Using the Snowflake Model Registry — Logging the model
• Using the Snowflake Model Registry — Calling the model

Section summary: This section highlights how Snowflake supports external model integration. By running third-party models with Snowpark Container Services or registering them in Snowflake’s Model Registry, you can extend AI beyond managed services to incorporate custom environments and external assets.

You will also explore fundamental technical steps like setting up compute pools and managing container images. This knowledge enables you to operationalize any model within Snowflake’s scalable framework.

Domain 3: Snowflake Gen AI Governance (22% of the exam)

Set up model access controls.

• Limits on which models can be used — Restrict access to specific models
• Limits on which models can be used — CORTEX_MODELS_ALLOWLIST parameter — Cortex LLM REST API
• Limits on which models can be used — CORTEX_MODELS_ALLOWLIST parameter — COMPLETE (SNOWFLAKE.CORTEX)
• Limits on which models can be used — CORTEX_MODELS_ALLOWLIST parameter — TRY_COMPLETE (SNOWFLAKE.CORTEX)
• Limits on which models can be used — CORTEX_MODELS_ALLOWLIST parameter — Cortex LLM Playground (Public Preview)
• Data safety and security considerations — Is data leaving/going to LLMs?
• REST API authentication methods

Section summary: In this section, you will gain the skills to implement access control for generative AI models. You will discover how to manage restrictions with configuration parameters and allowlists, ensuring that only sanctioned models are available for use within specific environments.

At the same time, the section emphasizes security and safety practices. This includes controlling data exposure and implementing authentication tools on REST APIs, protecting sensitive data while maintaining necessary model functionality.

Set guardrails to filter out harmful or unsafe LLM responses.

• Cortex Guard — COMPLETE arguments
• Methods to reduce model hallucinations and bias
• Error conditions

Section summary: Guardrails are crucial in any AI system, and this section equips you with best practices for deploying them in Snowflake. By working with Cortex Guard and applying filters for model completions, you can reduce risks while maintaining meaningful outputs.

You will also cover techniques for minimizing hallucinations and bias, along with strategies to account for error conditions. Together, these contribute to maintaining trustworthy generative AI applications.

Monitor and optimize Snowflake Cortex costs.

• Cortex Search — Different types of costs (virtual warehouse, EMBED_TEXT, Serving)
• Cortex Analyst — Snowflake Service Consumption Table
• Cortex LLM functions — Minimize tokens
• Cortex LLM functions — Token cost implications
• Tracking model usage and consumption — Usage quotas
• Tracking model usage and consumption — CORTEX_FUNCTIONS_USAGE_HISTORY view
• Tracking model usage and consumption — CORTEX_FUNCTIONS_QUERY_USAGE_HISTORY view

Section summary: Cost optimization is a key aspect of effective AI adoption. In this section, you will learn how to identify and manage expenses across Cortex Search, Analyst, and LLM functions, including token management strategies. Understanding consumption tables and usage history views strengthens your ability to measure cost efficiency.

By monitoring workloads closely and applying lean configurations, you create sustainable generative AI deployments. The tools introduced here ensure clarity and accountability while maximizing ROI from Snowflake AI services.

Use Snowflake AI observability tools.

• Snowflake AI observability (Public Preview) features — Evaluation metrics
• Snowflake AI observability (Public Preview) features — Comparisons
• Snowflake AI observability (Public Preview) features — Tracing
• Snowflake AI observability (Public Preview) features — Logging
• Snowflake AI observability (Public Preview) features — Event tables
• Implementation methods — Trulens SDK

Section summary: This section emphasizes evaluative and diagnostic tools for Gen AI observability. You will use features like evaluation metrics, logging, event tables, and tracing, which help track model outputs and system behavior. Aggregated comparisons support better model selection and calibration decisions.

You will also explore integrations like the Trulens SDK, which allow for extended functionality. Strong observability ensures AI applications remain transparent, reliable, and continually improving.

Domain 4: Snowflake Document AI (12% of the exam)

Set up Document AI.

• Virtual warehouse, database, and schema considerations
• Roles and privileges — USAGE
• Roles and privileges — OPERATE
• Roles and privileges — CREATE SNOWFLAKE.ML.DOCUMENT_INTELLIGENCE
• Roles and privileges — CREATE MODEL

Section summary: Document AI requires technical preparation, and this section walks you through it step by step. You will explore environment setup such as warehouses and schemas, alongside the roles and privileges that determine who can perform key operations.

By mastering this setup process, you ensure that your workspace is ready to train, deploy, and run AI document intelligence models. Correct configurations lay the foundation for accurate and efficient document analysis.

Prepare documents for Document AI.

• Upload documents
• Train the model
• Requirements (for example, formats, size limits)
• Question optimization best practices

Section summary: This section focuses on orchestrating model training workflows from uploaded documents. You will learn the acceptable requirements, formats, and optimization practices for ensuring high performance during training.

Topics such as question optimization are emphasized, helping you refine how Document AI delivers insights based on user prompts and context. By aligning document preparation with these best practices, you sharpen both accuracy and efficiency.

Extract values from documents using Document AI.

• Conditions
• <model_build_name>!PREDICT query
• Automation of data pipelines

Section summary: Here you will learn how to query with Document AI models to extract structured values from documents. This includes understanding how conditions work alongside the specialized predict query syntax.

Using automation in data pipelines ensures these operations scale across entire workloads. By applying these skills, document-based intelligence becomes a seamlessly repeatable part of your enterprise workflows.

Troubleshoot Document AI given a use case.

• Extracting query errors
• GET_PRESIGNED_URL function
• Requirements and privileges
• Cost and best practices considerations

Section summary: Finally, you will explore troubleshooting for Document AI. This includes handling query errors, applying functions such as GET_PRESIGNED_URL, and reinforcing the privilege requirements for smooth execution.

Practical guidance on optimizing costs and applying operational best practices ensures Document AI can be run reliably and sustainably. This completes your skillset for deploying Snowflake’s document intelligence features in the field.