AI Marketplace Docs

Architecture overview

1.1 Overview

Figure 1: AI app architecture overview

With a basic AI problem: AI service will require a minimum of 6 components (docker container). The number of components will increase when the AI ​​problem is complex and needs to be divided into multiple stages for processing

Examples for simple AI problems: Person detection, face detection, face embedding, face searching… For each problem like this, 6 components need to be developed and deployed

For more complex problems such as face recognition, it needs to be processed through many steps: face detection, then face embedding and finally searching. There are up to 3 stages in this problem, so there will be up to 3 AI services. Therefore, the number of components that need to be developed and deployed is 10 (add 2 components for each added service)

1.2 AudioCraft service

Audio Craft is an AI service that generates sounds and music based on user descriptions. It is divided into 2 sub-services that operate independently of each other. The AudioGen is used for generating sounds such as: car horns, cat meowing... and the MusicGen specializes in generating music (with melody). AudioCraft's system design is shown in the figure 2

Figure 2: AudioCraft architecture

Figure 3: Call API sequence diagram