Rômulo Ferreira Dos Santos, Jaime De Melo Gama Da Silva, Matheus Henrique de Souza, Paulo Cesar Rodrigues Borges
This article proposes an Artificial Intelligence model to support teaching in dynamic educational environments, characterized by student heterogeneity, curricular changes, and multiple modalities. The research addresses the gap in solutions that coherently integrate dynamic learning profiles, continuous progress monitoring, useful recommendations, and governance coupled with the teaching workflow. The study designs a modular, data-driven artifact evaluated in a hybrid scenario using LMS. The architecture operationalizes the cycle “data → inference → intervention → monitoring → replanning,” with layers of data (logs, assessments, and metadata), intelligence (inferences and profiles), intervention (recommendations to students and support for teacher decision-making), experience (dashboards/feedback), and trust (auditing, privacy, and bias mitigation). The core of the model combines risk/proficiency prediction, a recommendation engine based on behavioral signals, and pedagogical rules. To preserve teaching autonomy and reduce socio-ethical risks, it incorporates explainability and a human-in-the-loop module to parameterize, approve, and audit recommendations. The proposed evaluation integrates technical metrics (F1, AUC, Recall@K, NDCG@K, and calibration), pedagogical impact indicators (learning gains, engagement, and teaching load), and equity audits by subgroups, aiming at responsible and sustainable adoption.