GES-C01 Practice Exam — GES-C01：Specialty Gen AI

1. The question bank is cloud‑connected and updates automatically; no manual re‑acquisition is required.

2. Start practicing right after activating the question bank. It supports simultaneous use on websites and mini‑programs, with one‑click bilingual switching for each question.

3. Functions include online practice, mock tests, note‑taking, wrong‑question recording, etc., valid for one year.

4. Recommended practice order: Turn on review mode to browse questions → Complete sequential practice → Take mock exams for pre‑test self‑assessment.

5. Activation codes can be purchased by clicking Buy Now on the right or via our official Tmall flagship store.

6. For inquiries, contact customer service through mini‑program, WeChat, WhatsApp or LINE.

Exam information

1. Basic Exam Information (GES-C01)

- Exam Code: GES-C01

- Prerequisite: A valid SnowPro Core (COF-C02/COF-C03) or SnowPro Associate: Platform certification is required

- Exam Format: Online proctored exam via Pearson VUE or on-site exam at designated test centers

- Question Types: Single-choice questions, multiple-choice questions

- Number of Questions: Approximately 60 questions

- Exam Duration: 90 minutes

- Passing Score: 750 out of 1000 (roughly 75% accuracy)

- Exam Fee: $225 USD (taxes not included); $180 USD for candidates in India

- Languages: English only

- Validity Period: 2 years

- Recertification: Retake the current Generative AI Specialty exam or earn a higher-level Snowflake certification before expiration

2. Target Audience

- Data Engineers, Data Analysts, Data Warehouse Administrators

- Data Scientists, AI/ML Engineers

- Cloud Platform Architects, BI Developers, ETL Engineers

- Developers and architects building generative AI applications on Snowflake

- Technical professionals who plan to deploy generative AI workloads on the Snowflake platform

3. Exam Content & Weighting (GES-C01)

1. Overview of Snowflake Generative AI (26%)

- Architecture and core capabilities of the Snowflake AI Data Cloud

- Positioning of Gen AI components including Cortex, Document AI and Cortex Analyst

- Use cases and business value of generative AI in data cloud scenarios

- Snowflake Gen AI ecosystem: Notebooks, Streamlit and Model Registry

2. Core Capabilities of Snowflake Gen AI & LLMs (40%)

- Usage and optimization of Cortex LLM functions (COMPLETE, EMBED, CLASSIFY, etc.)

- Generative AI workflow design: RAG, prompt engineering and model invocation

- Development of AI applications on Snowflake (Snowpark and Streamlit integration)

- Model management and deployment: Model Registry and Snowpark Container Services

- Implementation of vector databases and vector search

3. Governance & Security for Snowflake Gen AI (22%)

- RBAC permission design for generative AI workloads

- Data privacy and compliance: data masking, access policies and AI ethics

- Model observability and monitoring

- Security best practices for generative AI applications

4. Snowflake Document AI (12%)

- Architecture and core features of Document AI

- Unstructured data processing: document parsing, OCR and information extraction

- Configuration, usage and integration scenarios of Document AI

- Design and optimization of document processing workflows

4. Registration & Exam Procedures

1. Account Registration

- Visit the official Snowflake Certification Portal: https://learn.snowflake.com/en/certifications/

- Register or log in to your account → Select SnowPro Specialty: Gen AI to proceed with registration

2. Exam Scheduling

- You will be redirected to Pearson VUE. Choose either online proctoring or an on-site test center.

- Complete payment of $225 USD (Credit card and PayPal are accepted).

3. Requirements for Online Exams

- Take the exam in a quiet, private room with no other people present.

- Device requirement: Windows or macOS computer with a working webcam and microphone.

- Install the proctoring software and complete ID verification (Passport or Driver's License).

5. Exam Preparation Recommendations

Official Resources

- Snowflake Official Documentation: https://docs.snowflake.com/

- Free Level Up online courses

- Official exam guide for SnowPro Specialty: Generative AI

Recommended Experience

- 1+ year of hands-on experience with Snowflake

- Proficiency in ANSI SQL

- Basic knowledge of generative AI concepts

6. Certification & Career Value

- You will receive an e-certificate and digital badge within 1 to 3 business days upon passing the exam.

- Certification Value:

- A highly recognized specialty certification focused on generative AI in the cloud data industry

- A competitive advantage for roles such as Data Engineer and Cloud Architect

- Globally recognized credential that demonstrates professional competence in Snowflake Gen AI project delivery

Sample questions

GES-C01 · Q1

Question #1 What privileges must be granted to a user with the FNR_DEVELOP ROLE so the user can use Document AI on an existing pipeline? (Choose three.)

A.
GRANT DATABASE ROLE SNOWFLAKE.ML_USER TO ROLE FNR_DEVELOP_ROLE;
B.
GRANT DATABASE ROLE SNOWFLAKE.DOCUMENT_INTELLIGENCE_CREATOR TO ROLE FNR_DEVELOP_ROLE;
C.
GRANT ROLE SNOWFLAKE.DOCUMENT_INTELLIGENCE_CREATOR TO ROLE FNR_DEVELOP_ROLE;
D.
GRANT CREATE SNOWFLAKE.ML.DOCUMENT_INTELLIGENCE ON SCHEMA . TO ROLE FNR_DEVELOP_ROLE;
E.
GRANT CREATE SNOWFLAKE.MODELS.DOCUMENT_INTELLIGENCE ON SCHEMA .
F.
GRANT CREATE MODEL ON SCHEMA . TO ROLE FNR_DEVELOP_ROLE;

Answer: BCF

This question assesses knowledge of the access control framework for Snowflake Document AI (officially branded Document Intelligence), a core generative AI capability covered in the SnowPro Specialty Gen AI GES-C01 exam. To interact with an existing Document AI pipeline, a custom role requires three core permissions: predefined service access roles to use the managed Document Intelligence service, and a schema-level privilege to create and manage model artifacts associated with pipeline inference runs. The suggested answer of B, C, F correctly covers all three mandatory permissions, while incorrect options reference either unrelated roles for other Snowflake ML services or invalid, non-existent privilege definitions that do not exist in the Snowflake platform. Option Analysis: A. Incorrect. The SNOWFLAKE.ML_USER database role grants access to general Snowflake Cortex ML functions and model management features unrelated to Document Intelligence. It is not a required privilege to use Document AI pipelines, so this option is irrelevant to the scenario. B. Correct. SNOWFLAKE.DOCUMENT_INTELLIGENCE_CREATOR is a predefined database role in the SNOWFLAKE system database that grants baseline access to the Document Intelligence managed service APIs and core functionality, which is a mandatory prerequisite for all Document AI operations including interacting with existing pipelines. C. Correct. The SNOWFLAKE.DOCUMENT_INTELLIGENCE_CREATOR account-level role grants supplementary permissions to view, access, and run inference on existing Document Intelligence pipeline resources, complementing the database role grant in option B to enable full pipeline usage. D. Incorrect. There is no CREATE SNOWFLAKE.ML.DOCUMENT_INTELLIGENCE privilege defined in the Snowflake access control model. This is a distractor option with an invalid, non-existent privilege syntax. E. Incorrect. There is no CREATE SNOWFLAKE.MODELS.DOCUMENT_INTELLIGENCE privilege available in Snowflake. This option uses a fabricated privilege structure that does not align with official Snowflake access control rules for Document AI. F. Correct. The CREATE MODEL privilege on the target schema where the Document AI pipeline resides grants the role permission to run inference jobs on existing pipelines, store output model artifacts, and interact with pipeline-related resources in the schema. This schema-level permission is required alongside the service access roles to fully use Document AI pipelines. Key Concepts: 1. Document Intelligence Access Control Layer: Snowflake Document AI uses a two-layer permission model consisting of predefined system roles for access to the managed service, plus schema-level object privileges for managing pipeline-related resources. This layered model is a core Gen AI access control topic for the GES-C01 exam. 2. CREATE MODEL Privilege for Gen AI Workloads: The CREATE MODEL schema privilege is required for all generative AI and ML workloads that generate model artifacts, including Document AI inference runs, as it controls the ability to create and store model-related objects in a target schema. 3. Snowflake System Predefined Roles: Predefined roles hosted in the SNOWFLAKE system database provide controlled access to managed Snowflake services, and must be explicitly granted to custom roles to enable usage of specialized features like Document Intelligence. References: Snowflake Document Intelligence Access Control Requirements, SnowPro Specialty: Generative AI Exam Study Guide

GES-C01 · Q2

Question #2 When using the SNOWFLAKE.CORTEX.FINETUNE function, which column names must be included in the SQL statement in the training_data_query parameter? (Choose two.)

A.
BASE_MODEL
B.
COMPLETION
C.
INFERENCE
D.
PROMPT
E.
TRAINING

Answer: BD

The SNOWFLAKE.CORTEX.FINETUNE function enables supervised fine-tuning of pre-trained Snowflake Cortex large language models to adapt them to domain-specific use cases. The training_data_query parameter accepts a SQL SELECT query that returns the labeled training dataset for the fine-tuning process. For the function to process the training data correctly, the query result must include two mandatory columns that form the input-output pairs required for supervised learning: the input text the model will process, and the expected target output the model should learn to generate. These required columns are named PROMPT and COMPLETION respectively, so options B and D are the correct choices. Option Analysis: A. BASE_MODEL is a top-level input parameter passed directly to the FINETUNE function to specify which pre-trained Cortex model to fine-tune, it is not a column required in the training_data_query result, so this option is incorrect. B. COMPLETION is the required column that stores the expected target output corresponding to each input prompt in the training dataset. This labeled output is required for the fine-tuning process to adjust model weights to generate correct, use case-specific responses, so this option is correct. C. INFERENCE refers to the process of running a trained or fine-tuned model to generate outputs for new, unseen inputs, it is not a valid required column name for the training dataset used with the FINETUNE function, so this option is incorrect. D. PROMPT is the required column that stores the input text examples that the model will receive as queries during both fine-tuning and post-deployment inference. This input is a core required component of the labeled training pairs for supervised fine-tuning, so this option is correct. E. TRAINING refers to the overall process of adjusting model parameters using labeled data, it is not a required column in the dataset returned by the training_data_query parameter, so this option is incorrect. Key Concepts: 1. Cortex Fine-Tuning Training Dataset Schema: The SNOWFLAKE.CORTEX.FINETUNE function requires training data to follow a strict schema with two mandatory columns, PROMPT for input text and COMPLETION for expected target output, to support out-of-the-box supervised fine-tuning workflows. 2. Supervised Fine-Tuning (SFT) for LLMs: SFT is the fine-tuning approach supported by Snowflake Cortex, which relies on paired input and output labeled examples to adapt a base general-purpose LLM to perform specific domain or use case tasks with higher accuracy than the base model. 3. FINETUNE Function Parameter Requirements: The training_data_query parameter must return a dataset that adheres to the required column naming convention, as the function automatically maps these columns to the input and target fields for the fine-tuning pipeline without additional user configuration. References: Snowflake Cortex LLM Fine-Tuning User Guide, FINETUNE (Cortex) SQL Function Reference

GES-C01 · Q3

Question #3 Which strategy should be used to improve latency when using the SNOWFLAKE.CORTEX.COMPLETE function?

A.
Use the REST API when invoking inferences
B.
Use the SQL API when using a Snowpark session object
C.
Set the temperature parameter close to 1.
D.
Pass a JSON schema to the response_format parameter

Answer: B

The question asks for a strategy to reduce latency when calling the SNOWFLAKE.CORTEX.COMPLETE large language model function. The correct answer, B, leverages the internal Snowpark session's SQL API to invoke the COMPLETE function directly within Snowflake's infrastructure. This eliminates the overhead of external network round trips, cross-environment authentication, and payload serialization/deserialization required for external API calls, directly reducing end-to-end inference latency. This approach is a core recommended optimization for Cortex function invocations when working within Snowpark environments, as it keeps all request processing inside Snowflake's low-latency internal service mesh. Option Analysis: A. This option is incorrect. Invoking inferences via the public REST API requires external network calls to Snowflake's inference endpoints, adding significant round trip time, authentication overhead, and payload processing latency compared to internal invocations. This increases rather than improves latency. B. This option is correct. Using the SQL API from an active Snowpark session object executes the COMPLETE function call entirely within Snowflake's internal infrastructure. No external network transit is required, and the request uses the existing authenticated Snowpark session context, eliminating redundant authentication steps and reducing end-to-end latency for inference requests. C. This option is incorrect. The temperature parameter controls the randomness and creativity of the LLM output, with values closer to 1 producing more varied, creative responses, and values closer to 0 producing more deterministic, consistent responses. This parameter has no impact on inference processing time or latency. D. This option is incorrect. Passing a JSON schema to the response_format parameter forces the LLM to generate output that conforms to a specified structured format. This adds additional post-processing and inference constraints that increase, rather than reduce, inference latency. Key Concepts: 1. Cortex LLM Function Invocation Paths: Snowflake Cortex LLM functions like COMPLETE can be invoked either via native SQL (including from Snowpark sessions) running inside Snowflake's environment, or via external public APIs. Native SQL invocations have consistently lower latency due to the absence of external network overhead. 2. Snowpark Session Context Optimization: Active Snowpark sessions maintain an authenticated, persistent connection to Snowflake's internal services, allowing built-in functions including Cortex functions to be called directly without additional authentication or external network transit, reducing latency for repeated invocations. 3. Cortex Function Parameter Impact: Core LLM parameters such as temperature, top_p, and max_tokens control output characteristics (creativity, length, diversity) but do not affect inference latency, while parameters that add structured output constraints can increase processing time. References: Snowflake Cortex LLM Functions: COMPLETE, Optimize Cortex LLM Function Performance

GES-C01 · Q4

Question #4 A Gen AI Specialist is building a semantic model for Snowflake Cortex Analyst and needs to provide details about each field. One field contains complex data with high cardinality.Which technique will keep the semantic model concise?

A.
Include custom_instructions to truncate the field.
B.
Reference the search optimization service on this field.
C.
Reference a Snowflake Cortex Search Service on this field.
D.
Reference a Snowflake Corte Analyst Verified Query Repository (VQR) on this field

Answer: B

This question assesses knowledge of Snowflake Cortex Analyst semantic model best practices for handling high-cardinality complex fields while maintaining model conciseness. The Snowflake Search Optimization Service (SOS) creates a persistent, optimized access path for high-cardinality fields, including complex and semi-structured data, that are frequently used in filter, equality, or substring search operations. When SOS is referenced on the field in the semantic model, the specialist does not need to add verbose metadata, extensive sample value lists, or custom filtering logic to the model definition to support performant queries on the field. This directly reduces unnecessary bloat in the semantic model, fulfilling the conciseness requirement while ensuring Cortex Analyst can efficiently access and process the high-cardinality field as needed. Option Analysis: A. Incorrect. Adding custom_instructions to truncate the field modifies data output rather than optimizing the semantic model structure. Truncation risks losing critical data required for accurate analysis, and does not reduce redundant content in the semantic model itself, so it does not address the stated conciseness requirement. B. Correct. The Search Optimization Service is purpose-built to accelerate queries on high-cardinality and complex fields. Referencing SOS on the target field eliminates the need to include verbose field metadata, custom filtering rules, or exhaustive sample value lists in the semantic model, directly supporting model conciseness while maintaining query performance for Cortex Analyst. C. Incorrect. Snowflake Cortex Search Service is designed for building low-latency search functionality for retrieval-augmented generation (RAG) and end-user search applications, not for optimizing individual field definitions in Cortex Analyst semantic models. It is not referenced in semantic model field configurations to reduce model bloat, so it is irrelevant to this scenario. D. Incorrect. The Cortex Analyst Verified Query Repository (VQR) stores pre-approved, trusted SQL queries for common business questions, rather than optimizing individual field definitions in the semantic model. VQR applies to full query patterns, not high-cardinality field handling, so it does not address the requirement to keep the semantic model concise. Key Concepts: 1. Cortex Analyst Semantic Model Design Best Practices: Semantic models define data structures, relationships, and metadata to enable Cortex Analyst to generate accurate SQL. Conciseness is prioritized by leveraging native Snowflake optimization services instead of adding verbose custom logic or redundant metadata for complex fields. 2. Search Optimization Service (SOS) Use Cases: SOS creates optimized access paths for high-cardinality fields, complex semi-structured data, and fields frequently used in filter or search operations. Referencing SOS in semantic models eliminates the need for extended field metadata, reducing model bloat. 3. High-Cardinality Field Handling for Cortex Analyst: High-cardinality fields (fields with large volumes of unique values) require special handling to avoid overloading semantic models with unnecessary metadata. Native Snowflake optimization services are the recommended approach to maintain both model simplicity and query performance. References: Snowflake Cortex Analyst Semantic Model Guide, Snowflake Search Optimization Service Documentation, https://docs.snowflake.com/en/user-guide/search-optimization-service

GES-C01 · Q5

Question #5 A Gen AI Specialist needs to assess if the feedback from a recent survey was generally positive, negative, or neutral. The Specialist wants the assessment to be reported using either a floating-point number from –1 to +1 (inclusive) or an abject with a categories fieldWhich statements will achieve the desired results? (Choose two.)

A.
SNOWFLAKE.CORTEX.CLASSIFY_TEXT (SURVEY_TEXT)
B.
SNOWFLAKE.CORTEX.COMPLETE (SURVEY_TEXT)
C.
SNOWFLAKE.CORTEX.ENTITY_SENTIMENT (SURVEY_TEXT, [‘positive’, ‘negative’, ‘neutral'])
D.
SNOWFLAKE.CORTEX.FINETUNE (SURVEY_TEXT);
E.
SNOWFLAKE.CORTEX.SENTIMENT (SURVEY_TEXT)

Answer: CE

The scenario requires two valid output formats for survey feedback sentiment assessment: a floating-point score between -1 and +1, or a structured object with a categories field classifying text as positive, negative, or neutral. The two correct Cortex functions each fulfill one of these output requirements, are purpose-built for sentiment assessment tasks, and require no custom model training to run inference on unstructured survey text. This aligns with the Gen AI Specialty exam focus on selecting appropriate pre-built Cortex functions for common natural language processing use cases without unnecessary custom development. Option Analysis: A. SNOWFLAKE.CORTEX.CLASSIFY_TEXT is a general-purpose text classification function that requires labeled training data or a custom prompt to define classification tasks. It is not purpose-built for sentiment analysis and does not return the required -1 to 1 score or predefined sentiment category object by default, so this option is incorrect. B. SNOWFLAKE.CORTEX.COMPLETE is a generative LLM function that returns unstructured natural language text completions in response to a prompt. It does not produce the structured numeric score or category object required for this use case, so this option is incorrect. C. SNOWFLAKE.CORTEX.ENTITY_SENTIMENT when passed the list of sentiment categories ['positive', 'negative', 'neutral'] returns a structured object that includes a categories field with classification of the input text into the specified sentiment classes, which directly matches the second required output format in the scenario, so this option is correct. D. SNOWFLAKE.CORTEX.FINETUNE is a function used to fine-tune custom large language models on domain-specific training data. It is not used for running real-time inference to get sentiment scores for existing text, so this option is incorrect. E. SNOWFLAKE.CORTEX.SENTIMENT is a pre-built sentiment analysis function that returns a single floating-point score between -1 (most negative) and +1 (most positive) for input text, which directly matches the first required output format in the scenario, so this option is correct. Key Concepts: 1. Snowflake Cortex Pre-built NLP Functions: These are serverless, fully managed functions optimized for common natural language processing tasks including sentiment analysis, text classification, and entity recognition, eliminating the need for users to deploy or maintain custom ML models. 2. Cortex Function Output Schemas: Each Cortex LLM function has a defined output structure, ranging from scalar numeric values to structured JSON objects and unstructured text, and selecting the correct function for a use case requires matching the function output to the required data format. 3. Zero-shot Sentiment Classification: Snowflake Cortex supports zero-shot classification of text into user-specified categories without custom training data, allowing users to define custom sentiment classes and receive structured classification outputs directly. References: Snowflake Cortex Sentiment Function Documentation, Snowflake Cortex LLM Function Documentation, https://docs.snowflake.com/en/user-guide/snowflake-cortex/llm-functions

FAQ

How many practice questions are available for GES-C01?

This question bank includes 61 GES-C01 practice questions covering single and multiple choice, each with answers and explanations.

Are GES-C01 practice questions available in Chinese and English?

Yes, GES-C01 practice questions are provided in both Chinese and English.

Can I try GES-C01 practice questions for free?

Yes. Free sample questions are available on this page, and the full question bank is available after signing up on Zhangxuetu.

Try practice questions Get the question bank

GES-C01 Practice Exam — GES-C01：Specialty Gen AI

Exam information

Sample questions

FAQ

Related exams