COF-C03 Question 130
Single answerCortex SearchA retail company wants to add a natural-language product help feature to its customer support portal. The source data is a Snowflake table containing product manuals, troubleshooting notes, and model numbers. Users need low-latency semantic search over this content, and support agents also want to filter results by product line and language. The data changes daily as new documentation is loaded. Which approach best meets these requirements using Snowflake Cortex Search?
- A
Create a Cortex Search service on the documentation table, specify the text content to index, and include product line and language as filterable attributes so the service can be queried with both semantic search terms and metadata filters.
- B
Create a materialized view over the documentation table and use ILIKE predicates on the text columns, because Cortex Search does not support filtering on structured attributes such as product line or language.
- C
Store embeddings in a separate table and query them directly with standard SQL only, because Cortex Search cannot refresh when source data changes and is intended only for static datasets.
- D
Use Search Optimization Service on the documentation table as the primary solution for semantic similarity, because it automatically interprets natural-language meaning and returns ranked conceptual matches.
Show answer and explanation
Correct answer: A
Explanation
The best answer is to use a Cortex Search service configured on the documentation data with the main text content indexed for semantic retrieval and structured metadata exposed as filterable attributes. This is the practical design for a support portal that needs natural-language search plus filters such as product line and language. Cortex Search is built for low-latency retrieval over Snowflake data and supports ongoing synchronization with source data, which is important when documentation updates daily. In contrast, ILIKE and materialized views provide lexical matching rather than semantic understanding, and Search Optimization Service is a performance feature for supported query patterns, not a semantic search engine. Snowflake documentation and product guidance position Cortex Search as the managed way to build retrieval applications over enterprise data stored in Snowflake.
- A. Correct.
Correct. Cortex Search is designed for low-latency hybrid/semantic retrieval over Snowflake data. A Cortex Search service can be created on source data and configured with searchable text plus attributes that can be used as filters at query time. This directly fits the requirement for natural-language search over manuals and notes while allowing agents to narrow results by product line and language. Cortex Search services are intended to work with Snowflake-managed data refresh from the source, making them suitable for documentation that changes regularly.
- B. Incorrect.
Incorrect. A materialized view with ILIKE can support keyword-style pattern matching, but it does not provide semantic retrieval comparable to Cortex Search. The statement that Cortex Search does not support filtering on structured attributes is the misconception here; Cortex Search is specifically built to combine semantic retrieval with attribute-based filtering in practical applications.
- C. Incorrect.
Incorrect. While embeddings are related to semantic search concepts, the claim that Cortex Search cannot refresh when source data changes is false. Managing raw embeddings manually also adds operational complexity and does not best match the requirement to use Cortex Search for a production support portal. The distractor reflects a common misunderstanding that semantic search in Snowflake must always be built from custom embedding pipelines.
- D. Incorrect.
Incorrect. Search Optimization Service improves performance for selective queries and can help certain lookup patterns, but it is not the primary feature for semantic similarity or natural-language meaning. It does not replace Cortex Search for ranked semantic retrieval over unstructured text. This option confuses search-performance optimization for SQL predicates with AI-powered semantic search.