hybrid search

Hybrid search combines dense (semantic) and sparse (keyword) retrieval for optimal results. **Why hybrid?**: Dense excels at semantic similarity but may miss exact matches; sparse catches exact keywords but misses synonyms. Together they cover both cases. **Fusion methods**: Reciprocal Rank Fusion (RRF) - combine ranked lists, Linear combination - weighted scores from both methods, Cascaded - sparse first then dense rerank. **RRF formula**: score = Σ 1/(k + rank_i) across retrieval systems, k typically 60. **Implementation**: Run BM25 + vector search in parallel, merge results, optionally rerank with cross-encoder. **Score normalization**: Min-max scaling, z-score normalization before combination. **Weight tuning**: Domain-specific - technical docs may favor keyword, conversational queries favor semantic. **Production systems**: Elasticsearch with dense vectors, Vespa, Weaviate hybrid mode. **Results**: 10-20% improvement over single-method retrieval on benchmarks. **Best practices**: Start with equal weights, tune on validation set, consider query-dependent weighting for advanced systems.