A Gen AI Specialist is designing a Gen AI-driven data pipeline in Snowflake to meet these requirements:Download PDFs from a website once an hour Load the PDFs into an Internal stage Extract text from the PDFs and load the text into a Snowflake table Use prompt engineering to generate an opinion on the text based on a given persona Automate the pipeline to maximize efficiencyWhat Snowflake features will be required to create this data pipeline using stored procedures? (Select THREE).
Answer(s): B,C,E
A: Snowpipe ❌ (Incorrect):Snowpipe is used for continuous, event-based loading of files into Snowflake tables, typically from cloud storage like S3 or Azure Blob. In this scenario, PDFs are downloaded from a website using a stored procedure and loaded into an internal stage , so Snowpipe is not required.B: Streams and tasks ✅ (Correct):Streams and tasks are required to automate and orchestrate the pipeline. Tasks can schedule the stored procedures to run hourly, and streams can track changes in tables if incremental processing is needed.Together, they maximize efficiency and automation.C: External access integration ✅ (Correct):External access integration is required to allow Snowflake stored procedures to access external websites . This enables downloading PDFs from the website into Snowflake, which is a key requirement of the pipeline.D: Cortex SENTIMENT function ❌ (Incorrect):The SENTIMENT function is designed for sentiment analysis (positive, negative, neutral). The requirement is to generate an opinion based on a given persona , which requires a generative LLM function rather than sentiment scoring.E: Cortex TRY_COMPLETE function ✅ (Correct):CORTEX.TRY_COMPLETE is used for LLM-based text generation with prompt engineering. It fits the requirement to generate an opinion on the extracted text using a defined persona and is suitable for use inside stored procedures.F: Dynamic tables using incremental refreshes ❌ (Incorrect):Dynamic tables are useful for declarative data transformations and incremental refreshes, but they are not necessary for downloading files, invoking LLMs, or orchestrating a Gen AI pipeline using stored procedures.
Which command should be used to create an internal stage to store PDFs that will be used to make Snowflake Document AI?
Answer(s): A
A: CREATE STAGE doc_ai_stage; ✅ (Correct):This command creates an internal Snowflake stage, which is exactly what is required to store PDFs for use with Snowflake Document AI. Internal stages are managed by Snowflake and are the recommended option when storing documents directly inside Snowflake for processing.B: CREATE STAGE doc_ai_stage STORAGE INTEGRATION = S3_INT; ❌ (Incorrect):Using a storage integration indicates an external stage (such as Amazon S3). Document AI requires documents to be stored in an internal stage, so this option is not suitable.C: CREATE STAGE doc_ai_stage ENCRYPTION = (TYPE = 'SNOWFLAKE_FULL'); ❌ (Incorrect):SNOWFLAKE_FULL is not a valid encryption type for creating stages. This command would fail due to invalid syntax.D: CREATE STAGE doc_ai_stage ENCRYPTION = (TYPE = 'SNOWFLAKE_SSE'); ❌ (Incorrect):While SNOWFLAKE_SSE (server-side encryption) is valid, it is already the default for internal stages. Explicitly specifying it is unnecessary and not required for Document AI, making this option incorrect for the question.
A Gen AI Specialist is building an automated content generator in Snowflake to produce customized product descriptions. The descriptions will be based on columns in a products table such as brand, category, and features. Which Snowflake Cortex LLM function should be used to generate the descriptions?
Answer(s): D
A: PARSE_DOCUMENT ❌ (Incorrect):PARSE_DOCUMENT is used to extract structured text and data from documents such as PDFs or images. It is meant for document ingestion, not for generating new content like product descriptions.B: EMBED_TEXT_768 ❌ (Incorrect):EMBED_TEXT_768 generates vector embeddings for text, which are used for semantic search or similarity comparisons. It does not generate natural language content.C: SUMMARIZE ❌ (Incorrect):SUMMARIZE is used to condense existing text into a shorter version. It does not create new, customized content based on structured inputs like product attributes.D: COMPLETE ✅ (Correct):SNOWFLAKE.CORTEX.COMPLETE is the generative LLM function used to create new text. It is ideal for generating customized product descriptions based on inputs such as brand, category, and features.
A Gen AI Specialist is creating a chatbot to answer questions based on a company's historical sales data. The data is stored in Snowflake as CSV and PDF files. The Specialist wants to use the agentic programming paradigm and interact with Snowflake through a single API. Which combination of Snowflake features will support this design? (Select THREE).
Answer(s): A,C,F
A: Document AI ✅ (Correct):The sales data includes PDF files , which are unstructured. Document AI is required to extract text and structured information from PDFs so the chatbot can use that data when answering questions.B: Cortex LLM Playground ❌ (Incorrect):The LLM Playground is only a UI-based experimentation tool for testing prompts. It is not used in production systems, does not support agentic workflows, and does not provide a single API for interaction.C: Cortex Search ✅ (Correct):Cortex Search enables semantic and hybrid search across structured data (CSV-loaded tables) and unstructured data (extracted PDF text). This is essential for a chatbot using a retrieval-augmented generation (RAG) approach to answer questions based on historical sales data.D: Cortex Analyst ❌ (Incorrect):Cortex Analyst is designed for natural-language-to-SQL analytics (BI-style querying). It is not agentic and does not support orchestrated reasoning or a single unified API for chatbot behavior.E: Cortex Fine-tuning ❌ (Incorrect):Fine-tuning customizes model behavior through training, which is not required to build a chatbot that answers questions using existing company data.F: Cortex Agents ✅ (Correct):Cortex Agents support the agentic programming paradigm and allow interaction with Snowflake through a single API . Agents can orchestrate reasoning, call tools like Cortex Search, and generate responses—making them ideal for chatbot design.
Which fields must be included in the body of an API request in order to submit the request to the Snowflake Cortex Analyst REST API endpoint? (Select THREE).
Answer(s): B,E,F
A: Stream ❌ (Incorrect):The stream field is used in some LLM APIs to enable streaming responses, but it is not required to submit a request to the Snowflake Cortex Analyst REST API. A request can be processed without streaming enabled.B: Content ✅ (Correct):The content field is required because it contains the actual user question or message being sent to CortexAnalyst. This is the natural-language input that the analyst interprets and translates into insights or queries.C: Request ID ❌ (Incorrect):A request ID is not required in the request body. Snowflake can automatically generate identifiers for requests, so this field is optional and not mandatory.D: Suggestions ❌ (Incorrect):suggestions are part of the response returned by Cortex Analyst to guide follow-up questions. They are not required (or expected) in the request body.E: Role of the speaker ✅ (Correct):The role field (such as user or system ) is required to indicate who is providing the content . This helps CortexAnalyst correctly interpret the context of the message.F: Path to the semantic model YAML file ✅ (Correct):Cortex Analyst requires the path to the semantic model YAML file so it knows how to interpret the data schema, relationships, and metrics when answering natural-language questions.
A Gen AI Specialist wants to present a full set of predefined, answerable questions to end-users with Cortex Analyst. How should the Verified Query Repository (VQR) be configured? (Select TWO).
Answer(s): A,E
A: Set the use_as_onboarding_question flag to true for each desired question. ✅ (Correct):To present a predefined list of answerable questions to end users, the Verified Query Repository must explicitly mark those questions as onboarding questions. Setting use_as_onboarding_question = true tells Cortex Analyst to surface these questions as suggested or starter questions in the user experience.B: Add [Suggested Questions] in Markdown at the end of each desired question. ❌ (Incorrect):Cortex Analyst does not use Markdown annotations to identify onboarding or predefined questions. Suggested or onboarding behavior is controlled through configuration flags, not formatting.C: Use questions that are complete sentences that end in question marks. ❌ (Incorrect):While good grammar improves readability, Cortex Analyst does not require questions to end with a question mark in order to include them in the VQR or present them to users.D: Use the names of the physical tables and columns that are defined in the underlying data set in the SQL queries. ❌ (Incorrect):Cortex Analyst does not operate on physical table or column names when using a semantic model. Using physical names breaks the abstraction layer provided by the semantic model.E: Use the names of the logical tables and columns that are defined in the semantic model in the SQL queries. ✅ (Correct):The VQR must reference logical table and column names defined in the semantic model. This ensures Cortex Analyst can correctly interpret, validate, and surface the predefined questions to end users.
When using Snowflake Cortex, which design factors have the MOST impact on model performance per credit? (Select TWO).
Answer(s): B,C
A: The size of the data set ❌ (Incorrect):The size of the underlying data set does not directly impact model performance per credit in Snowflake Cortex. Cortex LLM usage is driven by tokens processed and inference behavior , not by how large the stored data set is.B: The complexity of the queries ✅ (Correct):Query complexity has a major impact on performance per credit. Longer prompts, more instructions, complex reasoning, and larger context windows increase token usage and computation, which directly affects how many credits are consumed for a given level of performance.C: The number of concurrent users ✅ (Correct):Concurrency significantly affects performance per credit. More concurrent users can increase contention and latency, reducing efficiency and increasing the credits required to maintain acceptable performance levels.D: The availability of compute resources ❌ (Incorrect):Compute availability is managed by Snowflake and abstracted away from the user when using Cortex. While it affects overall system behavior, it is not a primary design factor the user controls for optimizing performance per credit.E: The availability region where the model resides ❌ (Incorrect):The region primarily affects latency , not model performance per credit. Credit consumption is not significantly influenced by where the model is hosted.
Which consideration should be made when using the Snowflake Cortex COMPLETE function?
A: The COMPLETE function can only be used with the Snowflake Arctic model ❌ (Incorrect):The SNOWFLAKE.CORTEX.COMPLETE function supports multiple large language models, not just theSnowflake Arctic model. Users can choose from different available models depending on their use case.B: The COMPLETE function can only process single string prompts ❌ (Incorrect):The function can process more complex prompts, including structured inputs such as JSON. It is not limited to a single plain-text string.C: The COMPLETE function will retain the state between one call and the next ❌ (Incorrect):Each call to the COMPLETE function is stateless . It does not remember previous prompts or responses unless that context is explicitly included in the current prompt.D: The COMPLETE function will generate a response using a specified language model based on a given prompt ✅ (Correct):This is the key consideration when using COMPLETE . The function generates text by applying a selected language model to the provided prompt, making prompt design and model choice critical.
Post your Comments and Discuss Snowflake GES-C01 exam prep with other Community members:
Our website is free, but we have to fight against AI bots and content theft. We're sorry for the inconvenience caused by these security measures. You can access the rest of the GES-C01 content, but please register or login to continue.