Classification Algorithm for Bonus Tool

This master's thesis details the development and implementation of an automated text classification pipeline for the internal bonus nomination process at an engineering consultancy firm. The project addresses the limitations of the existing keyword-based system, which struggles with misclassifications, manual maintenance, and a lack of transparency. The proposed solution involves a full-cycle MLOps pipeline, leveraging a combination of real and synthetically generated data to train a classification model. The system utilizes transformer-based embeddings and a Multilayer Perceptron (MLP) model, with integrated Explainable AI (XAI) features to provide clear explanations for each classification, thus building user trust. While the new machine learning model did not immediately surpass the high accuracy of the current rule-based system on the limited available data, the project successfully established a scalable and maintainable architecture.