Why Database Vector Search Requires Rethinking Indexes

Podcast hosts Lucas and Luna explore the limitations of traditional B-tree and hash indexes when applied to vector search for AI workloads. They discuss a production case where a SaaS company saw query latency spike 200x after switching to vector similarity search without changing their indexing strategy. The episode explains why approximate nearest neighbor (ANN) indexes like HNSW and IVF differ fundamentally from exact-match indexes, how data distribution affects performance, and why pre-filtering with scalar fields can break recall. Listeners learn one concrete lesson: vector databases demand a shift from exact to approximate thinking, and mixing scalar filters requires careful co-optimization. #VectorSearch #DatabaseIndexes #ANN #HNSW #IVF #SimilaritySearch #AIWorkloads #QueryPerformance #IndexTuning #SaaS #ProductionFail #Recall #Latency #TechPodcast #FexingoBusiness #BusinessPodcast #DatabaseTech #DataEngineering Keep every episode free: buymeacoffee.com/fexingo