The Stack Overflow Podcast What (un)exactly do you mean by semantic search?
22 snips
May 5, 2026 
Brian O'Grady, Head of Field Research and Solutions Architecture at Qdrant, specializes in vector search, embeddings, and scalable search systems. He contrasts traditional Lucene text search with modern vector databases. He explains when exact-match search is preferable, the trade-offs of bolt-on vector indexes, Qdrant's portability and scaling features, and future work on video embeddings and local agent contexts.
AI Snips
Chapters
Transcript
Episode notes
When Lucene Beats Vector Search
- Lucene excels at exact text search and analytics workloads like logs and security where precise matches matter.
- Brian O'Grady contrasts this with vector search which is approximate and loses exactness, so Lucene remains the right tool for exact-match scenarios.
Semantic Search Surfaces Related Results
- Vector search surfacing non-exact but semantically related results is valuable for user-facing discovery like e-commerce.
- Brian explains embeddings capture relatedness so searching 'iPhone' can also surface similar phones beyond exact text matches.
Don't Bolt-On Vector Search at Scale
- Avoid bolting vector indexes onto systems not built for them at scale because memory and latency can explode.
- Brian warns PG Vector and elastic bolt-ons work for prototypes but often force migration to a dedicated vector service at tens of millions of rows.