Architecture decisions for database

For building AI applications, there needs to be various databases for various requirements. Review the database architecture decisions for the speech and vision recognition with RAG AI pattern design based on functional aspects:

Architecture decisions for databases
Architecture decision Requirement Alternative Decision Rationale
Relational datastore Provide a highly available backend database that meets availability, performance, and resiliency requirements for the application
  • A database that is deployed on VSI and self-managed
  • Database for PostgreSQL
  • IBM® Cloudant® for IBM Cloud®
  • SQL server
Database for PostgreSQL A managed database that is highly available to store records across microservices. For this application, it uses IBM Cloud Databases for PostgreSQL.
Document datastore Provide a document-based database optimized to store unstructured data for bespoke front end application
  • Databases for PostgreSQL
  • Databases for MongoDB
  • Databases for Redis
  • IBM Cloudant
IBM Cloudant IBM Cloudant is available as an IBM Cloud service with a 99.99% SLA. IBM Cloudant elastically scales throughput and storage, and its API and replication protocols are compatible with Apache CouchDB for hybrid or multicloud architectures.
Vector datastore Provide a highly scalable vector data for Retrieval-Augmented Generation (RAG) application
  • Pinecone
  • watsonx.data Milvus
  • ChromaDB
watsonx.data Milvus Milvus is a vector database that stores, indexes, and manages massive embedding vectors that are developed by deep neural networks and other machine learning (ML) models. It is developed to empower embedding similarity search and AI application. Milvus makes unstructured data search more accessible and consistent across various environments.
Vector Datastore for embeddings Provide a vector data store for embeddings for machine learning.
  • Pinecone
  • Elasticsearch
  • ChromaDB
Elasticsearch Elasticsearch is used as a Database to store vector representations also known as embeddings created by using machine learning algorithms