Upgrading AI Into the Production-ready Stage with Microservice-based MLOps
A multinational company providing a complete marketing automation SaaS platform was already familiar with Machine Learning and Artificial intelligence but was looking for a way to bring its AI algorithm to the next production-ready level.
Addepto turned out to be the best partner to do that.
Case Study Shortcut
Challenge
Lack of MLOps Infrastructure and Automation
The company’s Machine Learning models were developed as monolithic, standalone modules without any MLOps support. This resulted in manual execution, no versioning, no automated data pipelines, and limited ability to scale or retrain models efficiently.
Limited Production-Readiness of AI Models
Though the AI solutions were functional, they weren’t built with deployment, monitoring, or retraining in mind. This made it difficult to adapt the models to changing conditions (e.g., evolving email provider rules) or ensure consistent performance in live environments.
Bottlenecks in Model Maintenance and Iteration
Without a modular or microservice-based structure, updating or improving specific components of the AI system risked disrupting the entire workflow. This slowed development cycles and hindered the ability to introduce iterative improvements or retrain models on new data.
Goal
The goal was to:
- Rewrite the operating system to a separate CLI application
- Develop MVP for a platform that actively monitors the quality of the standard MLOps model
- Build a retraining process
- Publish the model to the production environment
Outcome
With MLOps, the company can actively monitor and optimize Machine Learning models in production.
By making adjustments as needed, managers can always be up-to-date with mail providers’ requirements, making the system efficient in delivering mail to end-users.
For example, the number of emails sent to a particular hosting provider never exceeds the allowable number, which prevents them from being considered SPAM. It helped the company to manage the risks and enabled them to scale up their capabilities to meet the needs of growing demand without sacrificing performance or reliability.
Before
- The company aimed to enhance their AI algorithm to a production-ready standard.
After
- MLOps enables the company to monitor and optimize Machine Learning models in production, ensuring efficient mail delivery to end-users by keeping up with mail providers’ requirements.
Integrate those solutions in your company
Case Study Details
Approach
Refactoring AI Models for Production Scalability
- The Addepto team began by revamping the original AI models, which were built as single, tightly coupled modules.
Building a Microservice-Based MLOps Architecture
- Designed a modular MLOps platform using microservices, where each process (e.g., data preprocessing, model inference, monitoring) runs as an independent, replaceable service.
Implementing CLI-Based Workflow Control
- Rewrote the system’s core functions into a separate command-line interface (CLI) application, enabling controlled execution and easier automation of workflows.
- The CLI became the foundation for triggering model training, validation, and publishing steps within the MLOps pipeline.
Developing an MVP for Model Monitoring and Quality Assurance
- Built an MVP platform to actively monitor model performance, including precision, recall, and drift metrics.
- This allowed the client to track model degradation and ensure continuous compliance with mail provider standards and evolving spam detection algorithms.
Creating an Automated Model Retraining Pipeline
- Developed a retraining loop that can be triggered based on model performance thresholds or changes in data distribution.
- This ensured the system could self-adapt over time, keeping it aligned with business and technical requirements.
