AutoSort AI for Google Drive: A modular tool for thematic classification of teaching practices using generative AI

This article presents AutoSort AI for Google Drive, a modular, open-source software system that combines keyword based rules, semantic embeddings, and large language models for the automatic classification of educational documents, including lesson plans, pedagogical reflections, activity guides, and other related materials. Its workflow is based on retrieving files from Google Drive, extracting and preprocessing text, performing hybrid classification using keyword rules, embeddings, and LLMs, validating classification decisions, and automatically organizing documents in the cloud. AutoSort AI was implemented and evaluated using documents from the NTIC course at the University of Sonora. The results show that generative models outperformed traditional approaches: DeepSeek-chat-v3 achieved an F1-score of 0.9048, while Gemini-1.5-pro obtained the highest precision, with a value of 0.9167. Therefore, the software constitutes a useful tool for teachers, administrators, and researchers, as it enables users to select from different generative models for document classification and promotes the use of AI in educational settings.