Retrieval & Search
Techniques for finding and using relevant information with AI.
- RAG
A comprehensive guide to RAG - the dominant pattern for building AI applications that answer questions using your own data.
- Query Transformation
Improve retrieval by rewriting, expanding, or decomposing user queries before search — including query rewriting, multi-query retrieval, query decomposition, and related transformation techniques.
- Hybrid Search
Combine keyword and semantic search for more accurate and comprehensive information retrieval.
- Semantic Search
Learn how AI understands the meaning behind queries to find relevant results beyond keyword matching.
- Vector Search
Understand how vector databases find similar items using high-dimensional embedding comparisons.
- Vector Databases
Purpose-built databases for storing, indexing, and querying embedding vectors at scale — including managed options like Pinecone (Database + Nexus knowledge engine).
- Chunking Strategies
How to split documents for RAG - fixed-size, recursive, semantic, and document-aware chunking with production trade-offs.
- Embedding Models
Choosing and evaluating embedding models - OpenAI, Cohere, BGE, E5, and open-source alternatives for production RAG.
- Re-ranking
Cross-encoder rerankers that re-score retrieved documents for precision - the highest-ROI improvement in most RAG pipelines.
- Metadata Filtering
Pre-filter documents by metadata before vector search - tenant isolation, date ranges, document types, and access control.
- Vector Quantization
Compress embedding vectors with product, scalar, and binary quantization to cut memory and speed ANN search with controlled recall tradeoffs.
- ANN Indexes
Approximate nearest neighbor indexes - HNSW, IVF, IVF-PQ, DiskANN, ScaNN, and flat - that make vector search fast at scale.
- Late Interaction Retrieval
ColBERT-style multi-vector and token-level retrieval that scores query-document interactions at search time for higher precision.
- Retrieval Evaluation
Measuring retrieval quality - recall@k, MRR, nDCG, and building golden test sets for RAG pipelines.