Retrieval-Augmented Generation (RAG) · TRACK 03/05
Retrieval & Reranking
BM25 meets embeddings: hybrid search, RRF, rerankers, and query rewriting.
// THE TRACK
01 · START HERERetrieversYou'll understand the retriever's job in a RAG system and the difference between dense, sparse, and hybrid retrieval.BEGINNERRerankingYou'll understand how a reranker turns a broad set of retrieved candidates into a tight, high-precision shortlist — and when it's worth the extra latency.INTERMEDIATEBM25You'll understand how BM25 ranks documents by keywords and why it remains essential alongside embeddings in modern RAG.BEGINNERReciprocal Rank Fusion (RRF)You'll be able to explain the RRF formula and use it to merge ranked results from multiple retrievers.INTERMEDIATEHybrid SearchYou'll understand how hybrid search combines keyword and vector retrieval, and why it beats either approach alone.INTERMEDIATECross-Encoders vs Bi-EncodersYou'll understand the architectural difference between bi-encoders and cross-encoders, and why it forces the retrieve-then-rerank pattern.ADVANCEDMetadata Filtering in RAGYou'll learn how to attach and filter on metadata so RAG retrieves only documents from the right user, date range, or source.INTERMEDIATEQuery RewritingYou'll understand how rewriting, expanding, or splitting a user's question before retrieval improves which documents come back.INTERMEDIATEHyDEYou'll learn how HyDE generates a hypothetical answer and embeds it to find real documents that match the answer, not the question.INTERMEDIATECohere RerankYou will understand what Cohere Rerank is, how a managed reranking API improves a RAG pipeline as a second stage, and when to use it instead of self-hosting.INTERMEDIATEBGE RerankerYou will understand what the BGE reranker is, how a cross-encoder scores query-document pairs, and why it is the default self-hosted reranking choice.INTERMEDIATEColBERTYou will understand what ColBERT is, how late interaction with token-level embeddings differs from single-vector search, and why it can deliver high recall.ADVANCEDSPLADEYou will understand what SPLADE is, how learned sparse retrieval expands terms to bridge vocabulary gaps, and why it acts like a smarter BM25.ADVANCED