What is the primary function of the embedding model in a RAG system?

Question

Options

• ATo generate responses based on retrieved documents and user queries
• BTo encode queries and documents into vector representations for comparison
• CTo evaluate the faithfulness and relevance of generated Answers
• DTo store vector representations of documents and search for relevant passages

• A
  5% (3)
• B
  82% (46)
• C
  9% (5)
• D
  4% (2)

Explanation

B is correct because the embedding model's job is to convert text - both the user's query and the documents in the knowledge base - into dense numerical vectors (embeddings) that capture semantic meaning, enabling similarity comparison in vector space.

• A describes the LLM (generator), not the embedding model. The large language model takes the retrieved context and query to produce the final answer.
• C describes an evaluator/judge model, a separate component used in evaluation pipelines (e.g., RAGAS) to score output quality - not part of the core retrieval process.
• D describes the vector database (e.g., Pinecone, Weaviate), which stores embeddings and performs nearest-neighbor search. The embedding model creates the vectors; the vector DB hosts and queries them.

Memory tip: Think of the embedding model as a translator - it converts human language into a mathematical language (vectors) that machines can compare. The vector DB is the library that stores and searches those translations.

No community discussion yet for this question.