Email categorization is a significant challenge in today's digital workplace, where sorting through messages may become complicated. This thesis addresses this problem by developing and evaluating the Thunderbird email client's autonomous email categorization system.

Specifically, the study explores several machine learning models: Traditional algorithms (Logistic Regression, Support Vector Machines (SVM), Random Forest, XGBoost) and a Deep Learning model, Distil-BERT.

The study also showcases the development of a custom Thunderbird extension that integrates fine-tuned machine learning models through a FastAPI backend service, providing empirical verification of the proposed approach. This implementation supports single and ensemble model predictions, offering flexibility in dealing with the trade-off between accuracy and computational resource usage.

The findings of this research provide insights and lessons learned regarding the implementation, practical challenges, and future opportunities of automated email categorization, enriching the general understanding of using machine learning solutions for everyday productivity problems.