1. Introduction

In term of developed technical advancement, the amount of data increases significantly. Datas are diverse in the forms of text, picture, video and even from sensor data.

The challenge here is How can we store and retrieve data effectively ?

• Vector embeddings : The digital illustration of complex data such as text, image, sound and video.

The conventional method like RDBMS shows its weekness when it works with multimedia while NoSQL encounters difficulty with complicated retrival.

⇒ Vector Database becomes the characteristic method, handle effectively with similarity search, matching and recommendation which DBMS like SQL or NoSQL can not be optimal.

1. The conventional Database

SQL Database:

NoSQL:

1. Vector database

1. Definition

Vector database is a type of database designed exclusively for storage, index and retrieve vector embeddings

While the conventional databases demonstrates data in scalar form (number, series, data, …), Vector Database solves multidimensional data efficiently due to embeddings are being created - obtain feature for AI application like similarity search, semantic retrival.

First, we use an embedding model to create embedding vector for context we want to index.

Then the vector embedding is inserted to Vector database with indicated parameter to original vector.

When we apply a retrieval, we use Embedding Model to create Vector Embedding in order to retrieve and use these Vector embedding to retrieve Vector Database for finding vector embedding similarity. Then vector embedding similarity reindex to the original context.