DOMAIN-SPECIFIC LANGUAGE AND MODEL-DRIVEN ENVIRONMENT FOR ACCESSING DIVERSE VECTOR DATABASES

Abstract

In this paper, a model-driven environment for uniform access to various vector databases is presented. The environment supports connection to the target vector database, definition of data structures, manipulation of database objects, querying, and vector similarity search. The central element of the environment is the domain-specific language uniVEC, with a textual syntax. An important part of the environment is code generators, that transform models created by using uniVEC into executable programming code. The code generators generate Python scripts to support work with one of the chosen vector databases, such as: Pinecone, Milvus, Chroma, Weaviate, or Qdrant. The proposed environment aims to simplify work with vector databases, reduce learning time and support migrations between different vector databases.