In this paper, we demonstrate the system submitted to the shared task of medical text classification in Arabic. We proposed a single-model approach based on fine-tuned LLM-based embedding combined with hierarchical classical classifiers, achieving a competitive macro F1-score of 0.46 on the blind test set. We explored various modeling strategies, including tree-based ensembles, LLM, and hierarchical correction for rare classes, highlighting the effectiveness of domain-specific fine-tuning in low-resource settings. The results demonstrate that a single fine-tuned Arabic BERT variant can serve as a strong baseline in extreme imbalance scenarios, outperforming more complex ensembles in simplicity and reproducibility.

The rapid proliferation of AI-generated text has raised concerns. With the increasing prevalence of AI-generated content, concerns have grown regarding authenticity, authorship, and the spread of misinformation. Detecting such content accurately and efficiently has become a pressing challenge. In this study, we propose a simple yet effective system for classifying AI-generated versus human-written text. Rather than relying on complex or resource-intensive deep learning architectures, our approach leverages classical machine learning algorithms combined with the TF-IDF text representation technique. Evaluated on the M-DAIGT shared task dataset, our Support Vector Machine (SVM) based system achieved strong results, ranking second on the official leaderboard and demonstrating competitive performance across all evaluation metrics. These findings highlight the potential of traditional lightweight models to address modern challenges in text authenticity detection, particularly in low-resource or real-time applications where interpretability and efficiency are essential.