Scaling pgvector AI Workloads with YugabyteDB

Hello friends,
Our next Dallas–Fort Worth PostgreSQL Meetup is on Tuesday, August 11.

Event presenters: Dan Farrell, Solution Engineer/Architect at YugabyteDB

Abstract: pgvector turned Postgres into the default choice for AI and RAG applications—until scale hits back. Memory limits, recall degradation under metadata filters, and single-node ceilings force teams into painful re-architectures or a costly move to dedicated vector databases. This talk examines what actually breaks as vector workloads grow, and how a distributed, Postgres-compatible architecture from Yugabyte solves these issues without changing your SQL or your app code. We'll cover how YugabyteDB shards and distributes HNSW vector indexes across nodes, and share results from benchmarking 1 billion vectors with sub-second query latency. Live demos will show a RAG pipeline scaling from a single node to a distributed cluster in real time, including failover during active vector queries—something a single-node Postgres instance can't do. Finally, we will introduce Meko—Yugabyte's new Postgres-compatible agent-native data infrastructure for collective memory, shared knowledge and decision traces.

We'll have pizza, soft drinks, and social from 6:30 PM to 7:00 PM. The session will start at 7:00 PM.

Our generous sponsor, Improving, provides the place and pizza at every meeting. Thank you to everyone who RSVPs; this helps ensure we order enough food and reduce waste.
See you there!