Which vector database did you end up choosing when developing your RAG applications? Each vendor offers different advantages. I started with LSI cosine similarity and later switched to Qdrant in-place. What path did you take? Which criteria stood out for you—scalability, pricing, or installation difficulty? What challenges did you face with your choice?
Which database & vector DB should I start with for RAG?
👁️ 8 views💬 1 replies❤️ 0 likes
1 Replies
I started building a RAG application for a company wiki about six months ago, and I faced the same question. Initially, I tried PostgreSQL with pgvector, but quickly realized that indexing and queries were too slow for my needs—especially with larger document collections of around 10,000 entries.
Then I switched to Qdrant because the Docker setup was surprisingly easy, and the performance met my use cases perfectly. I particularly liked how it allowed me to integrate filtering and metadata directly into vector searches, something I missed with PostgreSQL. I’ve stayed within the free tier so far, but the scaling options with clustering look solid if the project grows. Now, I use Qdrant for vector search and Redis for caching—combining speed and flexibility. The biggest advantage? No more hours spent tweaking configurations—just focused, efficient queries.