Hybrid Search in PostgreSQL: Combining Vector and Full-Text for Real-World Applications
October 20–23
PostgreSQL is now a multi-model powerhouse, but many developers stop at using either pgvector for semantic search or tsvector for full-text. In reality, the best applications combine both. Semantic search understands meaning but misses keywords; full-text captures terms but ignores context. This session provides a practitioner's guide to building a unified hybrid search system.
We will dive into practical implementation patterns, covering schema design for embeddings and tsvectors, and the mechanics of Reciprocal Rank Fusion (RRF) using pure SQL, CTEs, and window functions. You'll learn how to execute parallel searches, combine disparate ranking systems, and apply metadata boosting for recency and popularity. Through live demos and query plan analysis, we'll explore how to avoid common pitfalls like score normalization errors. Leave with production-ready SQL patterns to build converged search workloads using nothing but PostgreSQL and community extensions.
