1.5 Describe Retrieval Augmented Generation (RAG) and Role of Embeddings and Vector Databases AI-TECHNICAL-PRACTITIONER Practice Questions (Page 4)

Questions 16–20

1. 16foundation · easy

   In a RAG system, how does an embedding model enable semantic search?

   • ABy ranking documents based on their word count and length
   • BBy assigning a unique ID to each document for lookup
   • CBy placing texts with similar meanings close together in a vector space
   • DBy matching exact keywords between the query and documents
   Reveal Solution Discussion Select an answer first
2. 17application · medium

   A RAG system for a medical Q&A bot retrieves chunks using cosine similarity. A user asks 'What are the side effects of ibuprofen?' The system returns a chunk about 'aspirin side effects' as the top result. What is the most likely reason for this incorrect retrieval?

   • AThe chunking strategy created a chunk that contains both ibuprofen and aspirin information.
   • BThe embedding model may not distinguish well between closely related drug names, so 'ibuprofen' and 'aspirin' are close in the vector space.
   • CThe LLM's generation stage rephrased the retrieved chunk to mention aspirin.
   • DThe vector database is using exact match search instead of semantic search.
   Reveal Solution Discussion Select an answer first
4. 18application · medium

   A legal firm wants to build an internal assistant that answers questions by citing specific clauses from its 50,000-document contract archive. The documents are stored as PDFs in a private SharePoint site. The team plans to use a pre-trained LLM that has no knowledge of these contracts. Which approach should they use to enable the assistant to answer questions with accurate citations?

   • AFine-tune the pre-trained LLM on the full text of all 50,000 contracts so it memorizes the clauses and can cite them from memory.
   • BUpload all PDFs directly into the LLM's context window at query time so the model can read every contract before answering.
   • CExtract text from the PDFs, split it into chunks, generate embeddings for each chunk, store them in a vector database, and retrieve relevant chunks at query time to include in the prompt.
   • DStore the PDFs in a relational database and use SQL keyword search to find documents containing the question's exact terms, then pass those documents to the LLM.
   Reveal Solution Discussion Select an answer first
5. 19foundation · easy

   What is the primary purpose of converting text into embeddings in a RAG system?

   • ATo translate the text into a different language
   • BTo represent the text as a vector that captures its semantic meaning for similarity search
   • CTo encrypt the text so that it can be stored securely
   • DTo compress the text so it takes up less storage space
   Reveal Solution Discussion Select an answer first
6. 20expert · hard

   A RAG system for a scientific research assistant retrieves chunks from a corpus of 2 million research papers. The team wants to improve the precision of retrieval—fewer irrelevant chunks in the top 5 results. Which approach is most likely to improve precision without significantly hurting recall?

   • AUse a hybrid retrieval approach: combine vector similarity with keyword (BM25) scoring and re-rank the merged results.
   • BSwitch to a vector database that uses exact nearest-neighbor search instead of approximate search.
   • CReduce the embedding dimension from 1536 to 128 to remove noise from the vectors.
   • DIncrease the number of retrieved chunks from 5 to 50 so the LLM can filter out irrelevant ones.
   Reveal Solution Discussion Select an answer first