From the course: LLM Foundations: Vector Databases for Caching and Retrieval Augmented Generation (RAG)

Choose a vector database

From the course: LLM Foundations: Vector Databases for Caching and Retrieval Augmented Generation (RAG)

Choose a vector database

“

Having discussed Milvus operations and two use cases for vector databases, let us now review some best practices for using vector databases. We begin with considerations for choosing a vector database. Vector databases is a fast-growing field that has not attained a mature state yet. We can expect them to evolve more in the near future. There are several options available for vector databases that makes the task of selecting one important. We have the option to choose a cloud service, or we can set up a standalone service in our data center. We can use an embedded in-memory database or deploy it in a scalable cluster. There are specialized vector databases as well as traditional databases that have vector support. The use case will determine the right vector database technology to use. So how do we decide on what type to choose for a given use case? We need to look at a few considerations. Do we need long-term storage? What are the scalability and reliability requirements? This will determine if we need a cluster or a cloud service. Do we have hybrid queries? If so, how frequent are they? This will determine whether we need to use a traditional database with vector support like PostgreSQL or go to a specialized vector database like Milvus. Another key question is with enterprise data. Is it okay for your organization to store data in the cloud, especially confidential ones? This is the key question to answer before deciding on cloud services. If we want to host the services ourselves, can we provide the people and hardware resources for deploying and managing these services on an ongoing basis? I strongly recommend doing some due diligence before deciding upon the right technology for the use case.

Contents