MariaDB 11 just introduced a vector database, and of course, I had to try it out. Spoiler: it’s awesome.

Recently, I needed to implement a RAG (Retrieval-Augmented Generation) system to power enhanced product search for an eCommerce platform. The idea was simple: let customers use natural language to find related products, instead of being limited by rigid keyword searches.

I decided to combine OpenAI embeddings with MariaDB’s new vector database—and it turned out to be a very cost-effective solution. The real challenge was making sure the embeddings stayed consistent, which meant building a chain of prompts to keep everything aligned.

Once that was in place, MariaDB’s cosine similarity function worked like a charm. The result: smarter product search, faster performance, and a surprisingly smooth integration.