Artificial Intelligence (AI) has made significant impacts on educational technology, yet building knowledge graphs that effectively support personalized learning remains a core challenge. In this work, we introduce InsightFlow, a graph-aware Knowledge Tracing (KT) approach driven by large language models (LLMs) and Generative AI (GenAI). InsightFlow generates customized learning pathways that adapt to a learner’s background, characteristics, and training goals. Unlike conventional KT models that focus narrowly on sequential prediction, InsightFlow integrates educator-curated knowledge graphs with Transformer-based architectures to produce high-fidelity mastery estimates over fine-grained knowledge components (KCs). These signals enable the construction of learning paths that are both highly personalized and, at the same time, aligned with specific training goals. In this paper, we also describe InsightFlow’s architecture, its graph-aware extensions, and its integration with GenAI models, establishing a scalable paradigm for goal-oriented, explainable, and user-defined personalized training.