1.5 Describe Retrieval Augmented Generation (RAG) and Role of Embeddings and Vector Databases AI-TECHNICAL-PRACTITIONER Practice Questions (Page 3)

Questions 11–15

1. 11application · medium

   A RAG system for a legal research tool retrieves the top 5 chunks from a vector database and passes them to the LLM. Users complain that answers sometimes miss important information that is spread across many documents. Which change is most likely to improve recall?

   • AIncrease the number of retrieved chunks (e.g., from 5 to 20) and ensure the LLM's context window can accommodate them.
   • BDisable the vector database and use the LLM's parametric memory instead.
   • CSwitch to a smaller embedding model so the vector database returns more diverse results.
   • DReduce the number of retrieved chunks to 1 to avoid confusing the LLM with too much information.
   Reveal Solution Discussion Select an answer first
2. 12foundation · easy

   Which scenario is best suited for using RAG instead of fine-tuning?

   • AWhen the knowledge base changes frequently and needs to be updated without retraining
   • BWhen the model must run offline without any external data access
   • CWhen the model needs to learn a new language from scratch
   • DWhen the model's response style needs to be permanently changed
   Reveal Solution Discussion Select an answer first
4. 13foundation · easy

   In a Retrieval Augmented Generation (RAG) pipeline, what is the primary function of the retrieval stage?

   • ATo convert the user's query into a numerical vector representation
   • BTo select relevant documents or text snippets from a knowledge source based on the user's query
   • CTo fine-tune the underlying language model on domain-specific data
   • DTo generate a natural language response based on the user's query
   Reveal Solution Discussion Select an answer first
5. 14application · medium

   A financial firm needs its LLM-based assistant to answer questions about its internal investment policy, which changes quarterly. The policy is a 200-page PDF. The firm wants to avoid retraining the model. Which approach best meets the requirement?

   • AStore the policy in the LLM's system prompt permanently.
   • BUse RAG: embed the policy PDF, store the vectors, and retrieve relevant sections at query time.
   • CFine-tune the model on the policy PDF each quarter.
   • DRetrain the model from scratch on the policy each quarter.
   Reveal Solution Discussion Select an answer first
6. 15application · medium

   A support team is building a RAG system over its troubleshooting guides. A user asks: 'My laptop won't turn on.' The retrieval step returns a guide about 'battery charging issues' even though the words 'laptop', 'turn on', and 'battery' do not appear in the query. Which component of the RAG pipeline is most directly responsible for this behavior?

   • AThe embedding model, which maps both the query and the documents into a shared vector space where semantically related text is close together.
   • BThe chunking strategy, which splits documents into fixed-size pieces before embedding.
   • CThe vector database's indexing algorithm, which organizes vectors for fast approximate search.
   • DThe LLM's generation stage, which rephrases the retrieved document to match the query.
   Reveal Solution Discussion Select an answer first