Ilias – Machine learning, Python, MLOps

A scalable machine learning model serving infrastructure on Kubernetes using KServe.

• Implemented scalable and fast machine-learning microservices;
• Coordinated with stakeholders and other researchers that implemented models in order to deploy them on the platform.

Balancing demand and supply with time series prediction and classification algorithm for many cities. Training and managing multiple models using Kubeflow pipelines and ArgoCD. Fully blown MLOps stack with integrated CI/CD and frequent releases. Model monitoring using Prometheus metrics and Grafana.

• Designed and implemented everything related to the project together with a team of ppl;
• Trained and deployed multiple models;
• Coordinated with stakeholders about the requirements and the fine-tuning of the system;
• Managed the team and resolved issues to enhance the workflow, particularly in testing and release processes.

Implementation of outlier detection in real-time drilling data with the aim of identifying problems/irregularities during drilling. Presently, costs due to drilling problems amount to hundreds of millions of dollars for the industry on a yearly basis. Used outlier detection methods like Isolation Forests to tackle the problem in a time-considerate manner effectively. Used data visualization and dimensionality reduction techniques (PCA, t-SNE) to allow a human-in-the-loop decision on ambiguous data points.

• Implemented the project end-to-end;
• Discussed and understood the problem with experienced drilling engineers;
• Extracted, transformed, and loaded (ETL) the required data;
• Conducted extended data visualization.

This project focuses on predictive analytics to identify customers with a higher likelihood of churning in the upcoming month. The primary objective is to leverage this predictive analysis to create personalized offers, effectively mitigating churn and bolstering customer retention.

• Conducted feature engineering;
• Collaborated with Campaign Managers to analyze requirements;
• Performed extended data visualization;
• Developed and trained models;
• Backtracked results and analyzed the performance of the model.