Snowflake SnowPro® Specialty: Gen AI Certification Sample Questions:

1. A company is using Snowflake AI Observability to evaluate a summarization application. The application utilizes SNOWFLAKE. CORTEX. COMPLETE for LLM inference and a custom Python component for text pre-processing. The team is particularly interested in tracking detailed cost breakdowns and assessing the factual correctness of the LLM-generated summaries. Which of the following statements accurately describe the cost implications and evaluation metric capabilities in this scenario?

A) The cost of AI Observability is primarily determined by the number of messages processed, and the number of tokens in each message does not affect the cost, ensuring predictable pricing.
B) AI Observability incurs charges for the LLM judges invoked via COMPLETE (SNOWFLAKE .CORTEX) calls to compute evaluation metrics, in addition to warehouse charges for managing runs and queries.
C) To measure the factual correctness of LLM-generated summaries based on original input and avoid hallucinations, the 'factual correctness' and 'comprehensiveness' metrics can be used during evaluations.
D) For evaluating summarization tasks, the 'context relevance' score is the most important metric, as it directly assesses the quality of the LLM's output against the source document.
E) The ORTEX_DOCUMENT_PROCESSING_USAGE_HISTORY view is the primary tool to monitor the credit consumption specifically for AI Observability evaluations and LLM judge usage.

2. A financial analytics team is using to extract specific financial metrics (e.g., revenue, profit margin) from quarterly reports and requires the output in a strict JSON format for automated ingestion into a data warehouse. They've encountered issues where the LLM sometimes generates malformed JSON or includes extraneous text. Which of the following approaches will help ensure deterministic, schema-compliant JSON outputs and mitigate these 'hallucinations' related to format?

A) Option E
B) Option B
C) Option C
D) Option A
E) Option D

3. A data engineering team wants to deploy a proprietary PyCaret classification model, saved as pycaret_best_model.pkl, for inference within Snowpark Container Services (SPCS). They need to register this custom model in the Snowflake Model Registry. Which of the following is a correct and essential step in this process?

A) Option E
B) Option B
C) Option C
D) Option A
E) Option D

4. A data engineering team is planning to build a real-time data pipeline using Snowflake's dynamic tables to process incoming log dat a. They want to use SNOWFLAKE. CORTEX. EXTRACT_ANSWER to pull out specific error codes and timestamps from log entries. They are also mindful of the operational costs. Which of the following statements accurately describes limitations or cost considerations for using SNOWFLAKE . CORTEX. EXTRACT_ANSWER in this scenario?

A) EXTRACT_ANSWER is explicitly designed for multi-language extraction, making it suitable for logs that might contain various languages without affecting cost.
B) The billing for EXTRACT_ANSWER is based on the combined token count of both the source_document (log entry) and the question used for extraction.
C) If a source_document input exceeds the model's token limit, EXTRACT_ANSWER will automatically truncate the text, potentially impacting results but avoiding an error.
D) The EXTRACT_ANSWER function does not support dynamic tables, which will prevent its direct use in this pipeline design.
E) To optimize performance and reduce cost for large-scale log processing, it is recommended to execute EXTRACT_ANSWER queries on a Snowpark-optimized warehouse of at least 'LARGE' size.

5. A marketing analyst wants to quickly gauge the overall sentiment of customer feedback stored in a Snowflake table called CUSTOMER_FEEDBACK, which has a column FEEDBACK_TEXT. They decide to use the SNOWFLAKE .CORTEX.SENTIMENT function to process a review. Consider the following SQL query for a specific review:

Which of the following correctly describes the expected output format and interpretation of the sentiment_score for the given input?

A) The output will be a JSON object containing a 'label' field, and values around 0 indicate a neutral sentiment.
B) The output will be a string like 'Positive' or 'Negative', and a score close to 1 indicates strong positive sentiment.
C) The output will be a boolean value (TRUE/FALSE) indicating if the sentiment is positive, and 0.5 represents a neutral sentiment.
D) The output will be a floating-point number between -1 and 1 (inclusive), where a value of 1 indicates strong positive sentiment and -1 indicates strong negative sentiment.
E) The output will be an integer between 0 and 1 00, where higher values denote more positive sentiment.

Solutions: