We present a Hybrid Graph Attention Model, a learning framework that integrates an efficient splitting algorithm applied to the Team Orienteering Problem (TOP) within a deep learning model. This hybrid approach operates in two steps. Initially, a giant tour (a sequence of customers/locations) is generated at once using a deep neural network. Subsequently, it is evaluated using the split algorithm. The primary objective is to narrow down the solution space in which the deep learning model operates, expecting an overall better performance. Two tailored solution approaches are employed to assess both the performance and the quality of the outcomes. Preliminary findings suggest that our method produces competitive solutions for the TOP.