179 questions
No questions match those filters.
What is Multi-Index RAG, and when does it actually earn...
This is one of the questions in the full AI/ML interview bank. Pro unlocks all 1789 questions; Premium includes the same bank plus the highest daily Practice limit.
See plansThe premise behind Multi-Index RAG is that not every kind of content retrieves well the same way, so forcing everything into one undifferentiated vector index means every content type gets a retrieval strategy that’s wrong for at least some of it. Instead, you maintain separate indexes tuned to what they hold: a summary index for document-level overviews that answers broad questions well, a chunk index for granular detail questions, a table index that handles structured financial or pricing data with a retrieval method suited to tables rather than prose, and potentially a knowledge-graph index for questions about relationships between entities.
A router — often an LLM doing simple classification — looks at the incoming query and decides which index (or indexes) it should hit: “what’s our vacation policy” routes to the text-chunk index, “give me an overview of the handbook” routes to the summary index, “what was Q3 EBITDA” routes to the table index. This earns its complexity specifically when a corpus is genuinely heterogeneous — mixing prose, tables, and broad-vs-specific query types — and the wrong answer, in that setting, is a single index trying to serve every retrieval pattern equally badly.