1.5 Describe Retrieval Augmented Generation (RAG) and Role of Embeddings and Vector Databases AI-TECHNICAL-PRACTITIONER Practice Questions (Page 2)

Questions 6–10

1. 6application · medium

   A healthcare company has a pre-trained LLM that writes patient education summaries. They need the model to adopt a specific tone and formatting style required by their compliance team. The medical knowledge itself is already well-covered by the pre-trained model and does not change frequently. Which approach is most appropriate?

   • AUse prompt engineering with a detailed style guide in the system prompt for every request.
   • BRetrain the model from scratch on the entire medical corpus with the new style.
   • CBuild a RAG pipeline that retrieves compliance documents and injects them into every prompt.
   • DFine-tune the model on a curated set of example summaries that demonstrate the required tone and formatting.
   Reveal Solution Discussion Select an answer first
2. 7expert · hard

   A multilingual company is building a RAG system where users ask questions in English, but the knowledge base contains documents in both English and Spanish. The team uses a single multilingual embedding model. They notice that English queries often fail to retrieve relevant Spanish documents. What is the most likely cause and best remedy?

   • AThe LLM's generation stage translates the query to Spanish before retrieval; remedy by disabling this feature.
   • BThe multilingual embedding model may not align English and Spanish vectors perfectly; remedy by using a model specifically trained for cross-lingual alignment or by translating queries/documents to a common language.
   • CThe chunking strategy splits Spanish text incorrectly; remedy by using a Spanish-specific tokenizer.
   • DThe vector database does not support multiple languages; remedy by using two separate vector databases, one per language.
   Reveal Solution Discussion Select an answer first
4. 8foundation · easy

   How does a vector database typically retrieve relevant documents for a query in a RAG system?

   • ABy using SQL queries to join tables based on foreign keys
   • BBy scanning every document for exact keyword matches
   • CBy performing a nearest-neighbor search on the query embedding
   • DBy using a decision tree to classify documents into categories
   Reveal Solution Discussion Select an answer first
5. 9application · medium

   A startup is building a RAG chatbot over its product manuals. The manuals are updated weekly. The team currently re-embeds every document and re-inserts all vectors into the vector database each week. They notice the process is slow and expensive. Which change would most directly reduce the cost of updating the index?

   • AOnly embed and insert vectors for documents that have changed since the last update, and delete vectors for removed documents.
   • BIncrease the chunk size so each document produces fewer vectors, even if chunks become very large.
   • CUse a smaller embedding model that produces lower-dimensional vectors, regardless of accuracy loss.
   • DSwitch from a vector database to a relational database that stores the raw text and uses LIKE queries.
   Reveal Solution Discussion Select an answer first
6. 10expert · hard

   A RAG system for a customer-support chatbot retrieves the top 3 chunks from a vector database. Users complain that answers are sometimes contradictory because the retrieved chunks contain conflicting information from different product versions. The team wants to ensure the chatbot always prefers the most recent product version. Which addition to the RAG pipeline is most effective?

   • AStore a version timestamp as metadata with each chunk, and filter or re-rank retrieved chunks by version before passing them to the LLM.
   • BIncrease the number of retrieved chunks from 3 to 10 so the LLM has more context to resolve conflicts itself.
   • CFine-tune the LLM to always prefer the most recent information in the prompt.
   • DRemove all older version chunks from the vector database so only the latest version is retrievable.
   Reveal Solution Discussion Select an answer first