Neon is a Postgres company. And we have always been fans of what ParadeDB built, especially the pg_search extension. So when we had the chance to work with them to offer this extension we were super excited. Not only can you easily index text for creating your own search engine but this takes AI RAG pipelines to a much higher level. When your RAG workflow goes to use any kind of text search you'll see much faster and better results thanks to the pg_search BM25 index.
If you don't know much about pg_search here's a breakdown:
pg_search is a high-performance full-text search extension for Postgres. It fills Postgres’ gaps in speed, ranking quality, and advanced search capabilities, bringing Elasticsearch-grade features inside Postgres.
The gaps are this. Postgres has native full-text search (FTS) via tsvector and tsquery, but this is not best-in-class—for example:
* Postgres’ GIN index on tsvector is not optimized for ranking, making queries slower when sorting by relevance.
* The built-in ranking method (ts_rank) is not as effective as modern search algorithms like BM25, which pg_search supports.
* Native FTS requires exact matches (no fuzzy search). pg_search can handle approximate matches and typos.
* Faceted search (the ability to quickly count occurrences of values within search results) is either missing or inefficient in native Postgres. This is a critical feature for real-world search applications and a major reason developers stick with Elasticsearch.
* Other advanced features like prefix search, highlight snippets, and JSON-aware search are also limited or inefficient in native Postgres FTS.
Replies
Neon