Free SnowPro® Specialty: Gen AI Certification Exam Practice Questions

Sample Practice Questions

Try these 10 sample questions from the SnowPro® Specialty: Gen AI Certification Exam practice bank. Each includes the answer and explanation. Continue with the interactive practice above to review the full question set.

1. 1What combination of retrieval methods underlies Cortex Search's hybrid search engine?

   1. A. Vector search combined with keyword search
   2. B. Keyword search combined with exact SQL joins
   3. C. Vector search combined with model fine-tuning
   4. D. SQL joins combined with model fine-tuning

   Answer: A

   Explanation: Snowflake describes a "hybrid (vector and keyword) search engine". Semantic and lexical retrieval are combined; fine-tuning and SQL joins are not the two retrieval components of this engine.

   Reference: Cortex Search overview — Overview

2. 2A text message can discuss several supplied topics, and AI_CLASSIFY must return every applicable label. Which config_object setting selects that mode?

   1. A. {'output_mode': 'single'}
   2. B. {'output_mode': 'multi'}
   3. C. {'task_description': 'multi'}
   4. D. {'task_description': 'single'}

   Answer: B

   Explanation: The output_mode documentation says: "Set to 'multi' for multi-label classification." Setting a task description does not select the output mode; single-label mode selects one label. This question uses text, avoiding the document-input restriction to single-label classification.

   Reference: AI_CLASSIFY — Arguments — config_object — output_mode

3. 3An administrator needs different account roles to access different Cortex models. Which approach is Snowflake's recommended model-access mechanism during the allowlist deprecation?

   1. A. Change CORTEX_MODELS_ALLOWLIST separately for each session.
   2. B. Give each role a different model-name list in its prompt.
   3. C. Set a separate CORTEX_MODELS_ALLOWLIST value on each user.
   4. D. Grant appropriate model application roles through RBAC.

   Answer: D

   Explanation: Snowflake recommends "migrating to role-based access control (RBAC) now", which provides per-role model control. The deprecated allowlist is account-level, and prompt text does not grant or restrict actual model permissions.

   Reference: Privileges and model access for Cortex AI Functions — Control model access — Account-level allowlist parameter — Warning

4. 4Which AI_PARSE_DOCUMENT options configuration extracts table structure along with the document's text?

   1. A. {'mode': 'OCR'}
   2. B. {'mode': 'LAYOUT'}
   3. C. {'mode': 'OCR', 'page_split': TRUE}
   4. D. {'mode': 'OCR', 'page_split': FALSE}

   Answer: B

   Explanation: LAYOUT "extracts layout as well as text, including structural content such as tables". OCR extracts text only. Setting page_split controls page segmentation; it does not turn OCR into layout extraction.

   Reference: AI_PARSE_DOCUMENT — Arguments — options — mode

5. 5Which description captures the core purpose of Cortex Analyst?

   1. A. Managing versions and dependencies of Python models
   2. B. Answering business questions over structured Snowflake data
   3. C. Building keyword and vector indexes over document text
   4. D. Scheduling containerized training jobs on GPU nodes

   Answer: B

   Explanation: Cortex Analyst helps create applications "capable of reliably answering business questions based on your structured". Its structured-data analysis role differs from registry management, document indexing, and container orchestration.

   Reference: Cortex Analyst — Overview

6. 6A review praises delivery but criticizes product quality. How should AI_SENTIMENT be configured to return separate sentiment for both aspects in addition to overall sentiment?

   1. A. Supply the review and a categories array containing delivery and quality.
   2. B. Supply the aspect names as input text instead of the original review.
   3. C. Append the aspect names to the review and omit the categories argument.
   4. D. Supply the original review as the sole argument without aspect categories.

   Answer: A

   Explanation: The categories argument is an "array containing up to ten categories" for which sentiment is extracted. Supply the original review and delivery/quality in that array to get aspect results along with overall sentiment. Mentioning aspect names in the input text or omitting the categories array does not request those separate results.

   Reference: AI_SENTIMENT — Arguments — categories

7. 7An account has the Snowflake-managed CORTEX-MODEL-ROLE-ALL bootstrap grant through SNOWFLAKE.PUBLIC. Which operation removes that bootstrap path before granting selected model roles?

   1. A. Revoke SELECT on every table in SNOWFLAKE.MODELS.
   2. B. Call SNOWFLAKE.LOCAL.REVOKE_FROM_PUBLIC_APPLICATION_ROLE for that application role.
   3. C. Revoke the model role directly from the account PUBLIC role only.
   4. D. Remove the model name from each application's prompt.

   Answer: B

   Explanation: Snowflake directs administrators to "Always use" SNOWFLAKE.LOCAL.REVOKE_FROM_PUBLIC_APPLICATION_ROLE for the bootstrap grant. A raw revoke from the account PUBLIC role does not remove the Snowflake-managed grant through SNOWFLAKE.PUBLIC.

   Reference: Privileges and model access for Cortex AI Functions — Default bootstrap of CORTEX-MODEL-ROLE-ALL — Important

8. 8AI_PARSE_DOCUMENT processes a one-page PDF with page_split set to TRUE. How is the extracted page represented in the parsed output?

   1. A. As a pages array containing one page object
   2. B. As an empty pages array because no split is needed
   3. C. As a pages array of strings without page objects
   4. D. As a top-level content field with no pages array

   Answer: A

   Explanation: Snowflake specifies a pages array "which contains a single object" even when the document has one page. The page_split option preserves the per-page object structure; it does not produce an empty array, an array of bare strings, or the unsplit top-level content structure.

   Reference: AI_PARSE_DOCUMENT — Returns — page_split output structure

9. 9A solution must plan a request, choose between structured-data analysis and document retrieval tools, and reason over their combined results. Which capability supplies this managed orchestration?

   1. A. Snowflake Model Registry
   2. B. Cortex Fine-tuning
   3. C. Cortex Agents
   4. D. Cortex Search

   Answer: C

   Explanation: An agent "reasons over a request, plans the work, calls tools, executes code, and generates a response". Cortex Agents provides the required orchestration; an individual retrieval service, tuning service, or model registry does not supply this combined workflow.

   Reference: Cortex Agents — Overview

10. 10When a SNOWFLAKE.CORTEX.TRY_COMPLETE operation cannot be performed, what does the function return?

    1. A. The unchanged input prompt
    2. B. A zero-length completion string
    3. C. The last successful completion
    4. D. A SQL NULL value

    Answer: D

    Explanation: TRY_COMPLETE "returns NULL instead of raising an error when the operation cannot be performed". An empty string, cached result, and original prompt are not its documented failure result.

    Reference: TRY_COMPLETE (SNOWFLAKE.CORTEX) — TRY_COMPLETE — description