Retrieval
Advanced retrieval strategies for RAG systems
Beyond basic vector search, modern RAG systems employ sophisticated retrieval strategies to handle complex queries and improve accuracy.
Core Concepts
Retrieval Fundamentals
Level: Beginner
Learn the core concepts of vector search and semantic retrieval including distance metrics, indexing strategies, and chunking.
Topics: vector-search • similarity-metrics • ANN-indexes • chunking-strategies
Reranking Fundamentals
Level: Intermediate
Improve retrieval accuracy by re-scoring candidates with a Cross-Encoder.
Topics: cross-encoder • bi-encoder • sentence-transformers • two-stage-retrieval
MMR: Maximal Marginal Relevance
Level: Intermediate
Diversify search results to reduce redundancy and improve coverage by balancing relevance with diversity.
Topics: result-diversification • redundancy-reduction • lambda-tuning • mmr-algorithm
Hybrid Search
Level: Intermediate
Combine semantic vector search with keyword-based BM25 search for more robust retrieval.
Topics: BM25 • semantic-search • score-fusion • reciprocal-rank-fusion
Query Expansion
Level: Intermediate
Improve recall by generating multiple variations of user queries to overcome vocabulary mismatch.
Topics: multi-query • query-decomposition • HyDE • step-back-prompting
Parent Document Retrieval
Level: Intermediate
Retrieve small chunks for precise search but return larger parent documents for better context.
Topics: parent-child-chunks • sentence-windows • hierarchical-retrieval • context-optimization
Advanced Techniques
Query Understanding and Classification
Level: Intermediate
Discover user query patterns using topic modeling and build classification systems to prioritize improvements.
Topics: BERTopic • persona-analysis • intent-classification • monitoring
Structured Metadata and Filtering
Level: Advanced
Combine semantic search with structured metadata filtering and SQL integration for constraint-based queries.
Topics: metadata-extraction • hybrid-retrieval • text-to-SQL • query-routing