Method of deploying an ai platform using MLOPS methodology

The specifics of modern AI applications require the development of a platform that enables continuous support and rapid reconfiguration. Scientific and engineering literature shows that the use of Machine Learning Operations and Machine Learning Pipeline (MLOps) methods when creating AI platforms for embedded systems can increase the efficiency of AI development and implementation. MLOps allows you to automate the development, training, and operation of machine learning models, providing a quick response to changes in the environment or input data. This helps to reduce the time for deployment and maintenance of the system as a whole. To solve the problem of deploying an AI platform, it is proposed to use a method based on a hybrid Edge-Cloud architecture. This approach is often used in the modern scientific literature and finds application in various fields. The architecture of the platform, based on hierarchical relationships between the Edge and Cloud layers, dictates the need to create a specialized approach to platform deployment. The peculiarity of this architecture is the use of embedded systems directly in AI tasks, namely, pattern recognition. This approach allows reconfiguring the platform, subject to updating the AI model, thereby maintaining its performance and correct task execution. Given that the key is to balance the speed of reconfiguration of such a platform with fault tolerance, in order to measure the overall performance of the system with a certain degree of accuracy, we will introduce several metrics that will help establish the parameters of the system and compare it with alternative solutions. Deployment of AI software should be accompanied by an analysis of the target system. As part of this work, a number of tests were conducted to quantify system performance.