Effective personalized programming education requires accurate assessment of a learner’s knowledge state to recommend optimal learning paths. However, traditional Knowledge Tracing (KT) models are ill-equipped to handle the intricate structural and semantic complexities of computer code. This limitation undermines existing Personalized Learning Path Planning (PLPP) frameworks, which depend on precise knowledge assessments and are further constrained by predefined knowledge structures (e.g., concept prerequisite graph). To address these challenges, we introduce an LLM-Aligned Cross-Hyperedge Tracer (LACHT), a novel framework that unifies KT and PLPP. LACHT models a learner’s code submissions as hypergraphs derived from Control Flow Graphs (CFGs), where each hyperedge encapsulates a distinct logical unit. A temporal tracing mechanism then monitors the evolution of these hypergraphs, weighting them by student performance to identify and select a set of key hyperedges that represent the learner’s most critical knowledge gaps. By pre-aligning hyperedge representations with an LLM’s semantic space, we enable the LLM to understand the knowledge concepts embedded within them. This allows the LLM to generate a pedagogically sound, personalized learning path based on the representations of the selected key hyperedges, critically eliminating the need for a predefined knowledge structure. Extensive experiments demonstrate that LACHT significantly outperforms state-of-the-art baselines in both knowledge tracing accuracy and the quality of generated learning paths.