I'm working on building an app similar to Duolingo and am considering using a vector database since I want to incorporate LLM (large language model) features in the future. I'm currently exploring pgvector, but since I've mostly used MySQL, PostgreSQL is pretty unfamiliar to me. I've heard that pgvector can have memory limitations and may be slow with large datasets. Considering my project's needs, is it a bad idea to start with a vector database right away? Or would it be better to stick with standard PostgreSQL for now and implement vector search later? Or should I go ahead with pgvector since I plan to use LLMs eventually? Any real-world advice would be super helpful!
3 Answers
I recommend starting with PostgreSQL for your app now. You can always add vector support later when you truly need it, and as long as your schema is well-designed, modifying it shouldn't be a problem.
You don't really need to pick just one! You could keep your relational data in PostgreSQL, and as you generate vector data, you can use a vector database alongside it. With pgvector, you can actually manage both types of data within a single database.
Vector databases are tailored for semantic search. Are you planning to use your LLM features for that purpose?
Yes, I plan to generate responses based on that! To be honest, I'm still learning about vector databases, so it's all a bit new to me.