Filter Extraneous Content in Source Documents That Degrades Quality of a RAG Application GENERATIVE-AI-ENGINEER-ASSOCIATE Practice Questions (Page 1)

Questions 1–5

1. 1expert · hard

   A financial services firm is building a RAG application over SEC filings (10-K, 10-Q). The documents contain extensive 'Risk Factors' sections that are largely boilerplate and repeated across filings, as well as unique 'Management's Discussion and Analysis' (MD&A) sections. The team applies a filtering step that removes all sections titled 'Risk Factors' to reduce noise. After evaluation, they find that while retrieval precision improved for queries about financial performance, the overall answer quality for queries about 'company-specific risks' dropped. The team needs to balance noise reduction with information preservation. What is the best course of action?

   • AKeep the Risk Factors sections removed and accept the lower answer quality for risk queries, as the improvement in financial performance queries is more valuable.
   • BRemove the entire Risk Factors section only for the most recent filing of each company, keeping it for older filings.
   • CUse a machine-learning classifier to identify and remove all sentences that mention 'risk' from the documents.
   • DReintroduce the Risk Factors sections but apply a more nuanced filter that removes only the boilerplate sentences repeated verbatim across multiple filings, preserving company-specific risk disclosures.
   Reveal Solution Discussion Select an answer first
2. 2application · medium

   A legal firm is building a RAG application over a large corpus of historical court rulings. The documents contain extensive boilerplate text such as 'IN THE COURT OF APPEALS OF THE STATE OF X', standard filing headers, and repetitive case caption information repeated on every page. The retrieval results are currently dominated by these repetitive elements, pushing substantive legal reasoning out of the top-k chunks. The team wants to remove this boilerplate while preserving the actual judicial analysis. What approach best addresses this problem?

   • AApply a regex-based pre-processing step that strips known boilerplate patterns (e.g., case captions, court headers) from each page before chunking.
   • BRemove all pages that contain the phrase 'COURT OF APPEALS' from the corpus before indexing.
   • CUse a machine-learning classifier to identify and remove all text that is not written by a judge.
   • DIncrease the chunk size so that each chunk contains more substantive text, diluting the impact of the boilerplate on retrieval.
   Reveal Solution Discussion Select an answer first
4. 3foundation · easy

   Which filtering approach is BEST suited to remove boilerplate text like 'For more information, visit our website' from a large corpus of PDF documents?

   • AA regex pattern that matches the exact phrase
   • BAn ML-based classifier that detects boilerplate language
   • CA heuristic that removes all sentences containing the word 'website'
   • DA simple length-based filter that removes short paragraphs
   Reveal Solution Discussion Select an answer first
5. 4foundation · easy

   A data engineer is preparing a corpus of technical manuals for a RAG application. Which type of content is MOST likely to degrade retrieval quality if left in the source documents?

   • ATables that contain structured specifications
   • BIntroductory paragraphs that provide an overview of the manual
   • CChapter titles that summarize the content of the following section
   • DPage headers and footers that repeat on every page
   Reveal Solution Discussion Select an answer first
6. 5foundation · easy

   To assess the impact of filtering on a RAG system, which evaluation approach is MOST comprehensive?

   • AReviewing the filtering code for correctness
   • BRunning a set of question-answering tasks and comparing answer quality before and after filtering
   • CAsking a domain expert to manually inspect a few filtered documents
   • DCounting the number of documents that were filtered out
   Reveal Solution Discussion Select an answer first