Deployment of machine learning services via microservices architecture

This work presents the design and implementation of microservices-based architectures for deploying machine learning models using Kubernetes. Two distinct architectures were developed: one focused on image classification using Convolutional Neural Networks (CNNs), and the other on a music recommendation system powered by a Large Language Model (LLM). Each architecture consists of specialized microservices, orchestrated via REST services and deployed as Docker containers managed by Kubernetes. Performance tests were conducted comparing these architectures against a monolithic implementation, evaluating metrics such as latency, scalability, and fault tolerance. The results showed that microservices-based architectures—especially the optimized version—achieved significant improvements in scalability and upgradeability, with a slight increase in latency. Three deployment strategies (Rolling Update, Blue-Green, and Recreate) were implemented to assess operational continuity. This study demonstrates that adopting microservices offers tangible benefits for flexible, scalable, and maintainable deployment of machine learning models in production environments.

• 第一著者: Zambrano Guanga, Steven Rafael (author)
• フォーマット: bachelorThesis
• 出版事項: 2025
• 主題: Microservicios; Aprendizaje automático; Microservices; Machine learning; Convolutional Neural Networks; Large Language Model
• オンライン･アクセス: https://repositorio.yachaytech.edu.ec/handle/123456789/994