Search on Percona Community

Building Smart Semantic Search using PostgreSQL and pgvector. Case Study - Part 2 - Postgres Layer

Sun, 31 May 2026 11:00:00 +0000

I’ll explain how I built the Postgres layer for semantic vector search on the Percona Community website: pgvector, chunks, two table modifications, the database schema, how the indexer populates Postgres, and what the SELECT statement looks like during a search.

Part 1: why semantic search, what’s already working on the site, the widget, and an overview of the stack.

Architecture

Search runs separately from the website at search.percona.community: FastAPI, a background indexer, and PostgreSQL with pgvector are all in a single Docker Compose file. percona.community remains static on Hugo and GitHub Pages, it doesn’t write directly to the database.

Building Smart Semantic Search using PostgreSQL and pgvector. Case Study - Part 1 - Introduction

Fri, 29 May 2026 11:00:00 +0000

Type “zero downtime database migration” into the site’s search bar and you’ll get articles and talks about database migration with minimal downtime, even if those words aren’t in the titles or content. This is semantic search on PostgreSQL and pgvector, without paid embedding APIs or a separate vector database. In this series I’ll cover how it works and why I chose this stack.

I’ll walk through how and why I built the search for our community site: blog, events, talks, and profiles. The post should help if you want to repeat the approach or need a practical case study on simple components. If you’ve done something similar, I’d like to hear your feedback.