Snowflake SnowPro Specialty Gen AI GES-C01 Dumps in PDF

Explanation:

Why option B is the best answer
The error message explicitly indicates a mime-type mismatch: “File extension does not match actual mime type. Mime-Type: application/octet-stream”. Document AI only accepts documents whose actual content-type matches an expected type (e.g., application/pdf , application/vnd.openxmlformats‑officedocument.wordprocessingml.document ).
When the content is generic application/octet-stream , the build fails with a processing error.
This situation occurs when the files staged for analysis were not uploaded with the correct MIME type (or the stage was created with an inappropriate format such as SNOWFLAKE_SSE ). Hence the most likely cause is that the source documents do not satisfy the Document AI ingestion requirements – option B.
Why the other options are less suitable
A – File URL expiration – An expired pre-signed URL would surface with authentication- or connectivity-related errors, not a MIME-type validation message. C – SNOWFLAKE_SSE encryption omission – Encryption settings affect security, not the MIME type reported by the stage; the error is unrelated to encryption. D – Incomplete statement (“The CREATE SNOWFLAK…”) – The option is cut off and does not reference any concrete issue; it also does not explain the MIME-type mismatch. E – Missing ML.DOCUMENT_INTELLIGENCE privilege – Privilege errors manifest as permission denied messages, whereas the reported error is a client-side processing validation, not an authorization failure.

Reference:

Staging files for Document AI: https://docs.snowflake.com/en/user-guide/data-share-prepare-stage Document AI requirements: https://docs.snowflake.com/en/snowpark-docs/document-ai/requirements

Explanation:

Why option A is required
Setting the use_as_onboarding_question flag to TRUE marks a query as part of the curated “starter” set that the VQR presents to end-users, ensuring the question appears in the predefined list that Snowflake surfaces during onboarding.
Why option E is required
VQR must reference logical tables and columns from the semantic model, not physical database objects. This guarantees that the query works across environments, respects row-level security policies, and automatically adapts if the underlying physical schema changes.
Why option C is required
Each question that will be stored in the VQR must be a complete sentence ending with a question mark . This enforces a uniform, natural-language interface that the Analyst can parse reliably and that users can read without ambiguity.
Why the other choices are not appropriate
B: Adding [Suggested Questions] in Markdown – Markdown is used only for documentation of the query itself; it does not influence how the VQR is loaded or displayed, so it has no effect on configuration.
D: Using physical table/column names – Queries that reference physical objects bypass the semantic layer, can break when the database schema evolves, and may violate security controls (e.g., hidden columns). The VQR expects logical identifiers that are abstracted by the model.
A + E together satisfy all technical constraints: they enable the curated onboarding list and guarantee that the underlying SQL leverages the logical schema that Snowflake’s Cortex Analyst understands.

Reference:

Verified Query Repository (VQR) configuration – Snowflake Documentation: https://docs.snowflake.com/en/user-guide/cortex-analyst/verified-query-repository
Semantic model best practices for Cortex Analyst – Snowflake Documentation: https://docs.snowflake.com/en/user-guide/cortex-analyst/semantic-model
These links provide official guidance on VQR setup and on using logical (semantic) objects in generated queries.

Explanation:

B: stream=True.
In many API implementations for Large Language Models (LLMs), enabling streaming is the standard way to receive responses in real-time as they are generated, rather than waiting for the entire response to be compiled and sent at once.
stream=True: This parameter instructs the server to send the response back as a series of "chunks" (often using Server-Sent Events or similar protocols). This significantly improves the user experience by reducing perceived latency, as the text begins appearing on the screen character-by-character or token-by-token.
Why the other options are incorrect:

A: deadline=5: This typically relates to setting a timeout threshold for a request, not the manner in which the response is transmitted.
C: options=CompleteOptions(max_tokens=5): This parameter limits the total length of the generated response to 5 tokens; it does not change the transmission mode.
D: options=CompleteOptions(response_format="incremental"): This is not a standard parameter in most widely used LLM SDKs (like OpenAI or Azure AI) for controlling streaming behavior.

Explanation:

Explanation:

Answer:
Event tables in Snowflake Cortex Analyst record every request, storing the natural-language question, the generated SQL, timestamps, and surrounding context, which enables auditing and debugging of model behavior. These logs allow analysts to trace how a particular output was produced and to review query patterns over time, supporting observability and troubleshooting. Option A describes archival of semantic models, a function handled by model versioning mechanisms, not by event tables. Option B refers to authentication logs, which are captured by security-related logging, not by Cortex event tables. Option C concerns warehouse performance metrics, which are monitored through separate query-history or performance-monitoring features.
References:
https://docs.snowflake.com/en/cortex/observability/event-tables https://docs.snowflake.com/en/cortex/analyst#observability

Reference:

References:
https://docs.snowflake.com/en/cortex/observability/event-tables https://docs.snowflake.com/en/cortex/analyst#observability

Explanation:

Technical Justification
The COMPLETE function in Snowflake Cortex is designed to invoke a specified language model and generate a response to a provided prompt. It therefore creates a model-driven output based on the input prompt , which aligns directly with option D . Option A is incorrect; COMPLETE works with any Snowflake-hosted model, not exclusively with Arctic mode. Option B is incorrect; the function can handle multi-token or multi-sentence prompts, not just single strings. Option C is incorrect; COMPLETE is stateless – each call is independent and does not retain context from previous invocations.
Conclusion: The only accurate statement is that the COMPLETE function generates a response using a specified language model based on a given prompt .

Reference:

1. Snowflake Cortex Documentation – COMPLETE function overview 2. Snowflake Cortex Overview – Languages & model invocation

Explanation:

Technical Justification
Cortex Search provides native vector-search capabilities built on Snowflake’s managed service, allowing users to index and query embeddings derived from unstructured data (e.g., PDFs, documents, logs). This capability is the foundation of Retrieval-Augmented Generation (RAG), where retrieved context is fed to an LLM to generate more accurate, knowledge-grounded responses. It is a production-grade feature designed specifically for semantic retrieval and integration with Snowflake’s data platform.
Cortex Analyst focuses on natural-language analytics over structured relational tables, converting user questions into SQL. It does not handle unstructured content or perform vector similarity retrieval, so it cannot directly power RAG pipelines.
Cortex LLM Playground is an interactive sandbox for experimenting with prompts and testing LLM integrations.
While useful for prototyping, it lacks the automated indexing, embedding generation, and secure, scalable retrieval workflow required for production RAG.
Semantic Views enable semantic search across tables by leveraging built-in search functions, but they are limited to structured metadata and do not provide the end-to-end RAG workflow that Cortex Search offers,
such as embedding storage, similarity filtering, and seamless LLM augmentation.
Therefore, B. Cortex Search is the only Snowflake feature that directly enables RAG using unstructured data in a production-ready manner.

Reference:

Cortex Search Documentation: https://docs.snowflake.com/en/user-guide/cortex-search Retrieval-Augmented Generation with Snowflake Cortex: https://docs.snowflake.com/en/developer-guide/cortex/rag-overview (or similar official Snowflake RAG guide)

Explanation:

The answer is B .
Technical justification The error Request failed for external function DOCUMENT_EXTRACT_FEATURESV1 with remote service error:
422 occurs when Snowflake’s external function times out or exceeds resource limits while retrieving a large number of documents. Option B directly addresses the root cause: limiting the volume of data processed or increasing the query timeout allows the external service to complete successfully.
Option A (Publish the model build) – Publishing a model does not affect the runtime limits of external function calls; it only registers the model for inference. Option C (Recreate internal stage with SNOWFLAKE_SSE encryption) – Encryption type only matters for stage security; it does not alter query execution constraints. Option D (Grant CREATE SNOWFLAKE privilege) – That privilege is unrelated to external function invocation and does not resolve the 422 timeout error. Option E (Grant ML.DOCUMENT_INTELLIGENCE privilege) – The privilege is required for accessing the service but does not mitigate the underlying quota or timeout issue that triggers the 422 response.
Hence, reducing document count or extending query expiration directly resolves the external function failure.

Reference:

Snowflake Documentation – External Functions Overview: https://docs.snowflake.com/en/sql-reference/external-functions Snowflake Documentation – Limits for External Functions and Services: https://docs.snowflake.com/en/sql-reference/external-functions-limitations#query-timeout-and-chunking