SnowPro Specialty: Gen AI Question 190
Single answerData enrichmentA retail company is building a support chatbot in Snowflake to answer questions about products, orders, and return policies. The team already stores raw support tickets, product catalog data, and policy documents in Snowflake. They want to improve response relevance by enriching unstructured ticket text so it can be used effectively in a retrieval pipeline for Cortex AI workloads. Which approach is the MOST appropriate for data enrichment in this scenario?
- A
Use Snowflake to chunk the ticket and policy text, generate vector embeddings for the chunks, and store the enriched records with metadata such as product category, region, and document source for retrieval.
- B
Export all support tickets to an external application, convert them into CSV files, and reload them into Snowflake because structured files are required before Gen AI features can use the data.
- C
Create a larger virtual warehouse so the chatbot can infer product and policy relationships at query time without adding metadata or embeddings to the source data.
- D
Duplicate the raw text into multiple tables by business domain so that the LLM has more copies of the information available during generation.
Show answer and explanation
Correct answer: A
Explanation
The best answer is to enrich the unstructured content so it can be effectively retrieved during generation. In Snowflake Gen AI patterns, this typically means transforming long documents into chunks, generating embeddings for semantic similarity search, and storing metadata that supports filtering and governance. This is a practical application of Retrieval-Augmented Generation (RAG), where retrieval quality strongly depends on how well the source data has been enriched. Snowflake documentation and best practices for Cortex AI and vector search emphasize preparing text into manageable chunks, generating embeddings, and preserving business metadata so downstream applications can retrieve relevant context efficiently and accurately.
- A. Correct.
Correct. For Gen AI retrieval scenarios in Snowflake, data enrichment commonly includes preparing unstructured text for search by chunking content, generating embeddings, and attaching useful metadata. This enriched representation improves retrieval quality for RAG-style applications by helping similarity search return the most relevant chunks and by enabling metadata filtering such as product line, geography, or source type.
- B. Incorrect.
Incorrect. Converting unstructured text to CSV does not meaningfully enrich the data for Gen AI use cases. Snowflake can work with unstructured and semi-structured content without requiring this detour. The misconception is that Gen AI workflows require conventional tabular reshaping first, when in practice enrichment focuses on making the content retrievable and context-aware.
- C. Incorrect.
Incorrect. Increasing warehouse size may improve performance, but it does not enrich the underlying data. Compute alone does not create semantic representations or retrieval metadata. The misconception is confusing infrastructure scaling with data preparation for Gen AI.
- D. Incorrect.
Incorrect. Copying the same raw text into multiple tables increases storage and management overhead but does not improve semantic retrieval. Without chunking, embeddings, or metadata, the chatbot still lacks the enriched context needed for accurate retrieval. The misconception is that more physical copies of data somehow improve model understanding.