Filter Extraneous Content in Source Documents That Degrades Quality of a RAG Application GENERATIVE-AI-ENGINEER-ASSOCIATE Practice Questions (Page 2)

Questions 6–10

1. 6foundation · easy

   When filtering extraneous content from source documents, what is the PRIMARY risk to RAG quality?

   • ALeaving too much extraneous content in the corpus
   • BRemoving content that is actually relevant to user queries
   • CIncreasing the storage cost of the vector database
   • DMaking the filtering process too slow
   Reveal Solution Discussion Select an answer first
2. 7expert · hard

   A multinational corporation is building a RAG application over its internal policy documents. The documents are available in multiple languages (English, Spanish, French). The team applies a filter to remove all non-English text to simplify the pipeline. After evaluation, they find that retrieval for queries in English has improved, but the answer generation for queries about policies that were originally written in Spanish has degraded, even when the query is in English. What is the most likely cause?

   • AThe filter was applied after chunking, causing the embeddings to be computed on the mixed-language text, which degraded the retrieval of the correct chunks.
   • BThe Spanish-language policy documents contained unique policy details that were not translated into English, and removing them eliminated the only source of this information.
   • CThe filter was too conservative and failed to remove all the Spanish text, leaving noise in the index.
   • DThe answer generator is not capable of translating Spanish text to English, so the degradation is unrelated to the filtering.
   Reveal Solution Discussion Select an answer first
4. 8application · medium

   A media company is building a RAG application over a collection of news articles. The articles contain author bios, 'Related Articles' links, and social media share buttons. These elements are consistently formatted but vary in content. The team wants to filter these elements to improve retrieval quality. Which approach is most appropriate?

   • AUse a combination of CSS selectors and regex to remove the HTML elements containing author bios, related article links, and share buttons before converting to plain text.
   • BApply a keyword filter to remove any text containing 'share', 'related', or 'author'.
   • CRemove any sentence that contains a person's name, as these are likely part of author bios.
   • DUse a general-purpose text summarization model to condense each article, removing any text that is not part of the main story.
   Reveal Solution Discussion Select an answer first
5. 9application · medium

   A legal tech startup is building a RAG application over case law documents. The documents contain a 'Procedural History' section that summarizes the case's journey through the courts, and a 'Discussion' section that contains the court's reasoning. The team applies a filter that removes the 'Procedural History' section to reduce noise, as it is often repetitive. After deployment, they find that queries about 'why the case was remanded' now return poor results. What is the most likely cause?

   • AThe filter was too conservative and failed to remove enough of the procedural history, leaving noise in the index.
   • BThe retrieval system is not configured to handle queries about 'remand', so the degradation is unrelated to the filtering.
   • CThe 'Procedural History' section contained the specific details about the remand, which were not repeated in the 'Discussion' section.
   • DThe filter was applied after chunking, causing the embeddings to be computed on the noisy text, which degraded the retrieval of the correct chunks.
   Reveal Solution Discussion Select an answer first
6. 10expert · hard

   A team is building a RAG application over a large corpus of technical support tickets. The tickets contain a 'Resolution' field with the final solution, and a 'Comments' field with back-and-forth discussion. The team applies a filter that removes the 'Comments' field to reduce noise, as it contains a lot of irrelevant chatter. After evaluation, they find that retrieval precision has improved, but the answer generation for queries about 'troubleshooting steps' has degraded. The team is considering whether to revert the filter. What is the best course of action?

   • AKeep the filter and accept the degradation, as the improvement in retrieval precision is more important.
   • BRevert the filter entirely, as the degradation in answer quality is unacceptable.
   • CRemove the filter only for tickets that contain the word 'troubleshooting' in the comments.
   • DKeep the filter but also apply a secondary filter to the 'Comments' field to remove only the irrelevant chatter (e.g., greetings, acknowledgements) while preserving the troubleshooting steps discussed.
   Reveal Solution Discussion Select an answer first