Educational success largely depends on learners' active engagement. Thanks to advances in artificial intelligence (AI) and the rise of online learning platforms, it is now possible to model and predict student engagement, offering more tailored learning pathways. Improving the learning experience is at the core of approaches aimed at optimizing pedagogical methods. This internship employs artificial intelligence (AI) technologies to analyze large-scale data from learning platforms, with a focus on studying learner engagement and behavior. It relies on the public EdNet dataset, which includes over 130 million student interactions with educational exercises, to identify predictive models of success and indicators of disengagement. The primary objective of the internship is to develop various approaches to modeling learner interactions (e.g., Knowledge Tracing models, recurrent neural networks, Transformer-based models) to assess their ability to predict the evolution of knowledge and estimate the level of student engagement. These models will be compared to existing methods and evaluated in terms of relevance, predictive accuracy, and their potential to support personalized learning pathways. The project will also investigate the relationships between interaction patterns (sequence of responses, response times, and success rates) and engagement levels, informing the development of adaptive pedagogical strategies.