Unified AI & Relational Power: Scalable Distributed Vector Search with
summary
YugabyteDB claims millisecond semantic search across billions of embeddings by sharding vector indexes with a Vector-LSM layer and running HNSW-powered local indexes merged across tablets for parallel top-K results. It combines PG vector storage, USearch and HNSW indexing, Raft-based shard consistency, automatic tablet splitting and geo distribution so vector and relational queries run as a single hybrid SQL+vector query. Demo loads a 5,000-row TMDB CSV with 384-D embeddings and filters results by rating > 7.5.
AI-written summary · about
chapters (9)
- 0:00Unified AI and relational data
- 2:30Relational vs vector data challenges
- 7:10Why separate systems fail
- 8:57Yugabyte DB overview
- 14:20Yugabyte DB for AI features
- 16:53Telecom plan recommendation demo
- 17:13Vector DBs vs Yugabyte DB
- 21:55Hybrid SQL and vector search demo
- 25:30Summary
related talks
33mBringing Vector Search to OLTP
36mUnleashing the Power of Vector Databases for AI-Driven Applications
47mBuilding intelligent analytics at scale
27mMachine Learning for Advanced AI Search
59mContextual search with vector search: exploring your options with open source tools
32mDistributed Vector Databases - What, Why, and How
35mVector Databases: Using Vector Search with Graph and Key-Value Queries
24mVector Search at Uber