Chunking, embeddings, vector search, reranking and the evaluation that tells you whether any of it worked.
10h · 8 lessons · 1 challenge
Before this: NLP Tasks & Production Pipelines, Prompting & Structured Output
RAG is the most-built and least-evaluated LLM architecture. This module takes it seriously as a retrieval problem first: chunking strategy, hybrid search, reranking, and separating retrieval failures from generation failures when the answer is wrong.
It is also the most commonly requested system design interview topic right now, so the module ends with that framing.
Wrong answers from a RAG system. Attribute each to chunking, embedding, retrieval depth, reranking or generation — then fix it and confirm.
Worth 600 XP.
