Generative AI Engineer is building a RAG application that answers questions about technology- related news articles. The source documents may contain a significant amount of irrelevant content, such…

Question

Options

• AInclude in the system prompt that the application is not supposed to answer any questions
• BFilter out irrelevant news articles in the retrieval process.
• CKeep all news articles because the RAG application needs to understand non-technological
• DFilter out irrelevant news articles in the upstream document database.

• A
  3% (1)
• B
  3% (1)
• C
  83% (24)
• D
  10% (3)

Explanation

Option C is NOT advisable because keeping all irrelevant articles (sports, entertainment, ads) in the vector store pollutes the retrieval index. When the retriever searches for technology-related context, these irrelevant chunks can rank highly and displace relevant tech content, degrading answer quality. The RAG system does not need to 'understand' unrelated domains to answer technology questions - that reasoning is flawed. Option A (system prompt instruction) is a valid lightweight guardrail. Option B (filtering in retrieval) is a valid runtime approach. Option D (filtering in the upstream document database) is the most effective approach as it prevents irrelevant data from ever entering the index, reducing storage and improving retrieval precision at the source.

No community discussion yet for this question.