QED-Tutrix (QEDX) is an intelligent tutoring system which assists students in proof problem solving by providing hints while taking into account the student's cognitive state. QEDX stands out by the fact that it adapts to each user and class reality, not the opposite. However, this model implies recognizing, like a teacher would, proofs that do not necessarily conform to a formal logic. Hence, QEDX can't rely on an automated proof engine (Tessier-Baillargeon, 2016), raising the question of how to expand QEDX's problem database without manually implementing each valid proof. Therefore, our poster at CERME10 doesn't present a traditional research project with its research questions, it's methodology and conclusions. It rather aims at presenting the new research questions that stem from the challenges that arise with trying to broaden QEDX's problem bank while staying true to our main goal, which is to create a geometrical workspace (Kuzniak, 2006) according to witnessed student/teacher interactions through a design in use approach (Rabardel, 1995). Here we will focus our attention on the process of problem implementation, starting with how we currently generate a proof problem's solution graph. Generating a proof problem's solution graph. QEDX's HPDIC graph (Figure 1) is used to record all the valid proofs to a given problem. It includes Hypotheses, Properties, Definitions, Intermediary results and a Conclusion. This graph is unique to each problem and is built from the inferences individually identified as true according to the problem to solve and the class context. The HPDIC graph for the rectangle problem in Figure 1 is fairly simple since it counts only 13 inferences. However, in the five problems implemented in the current QEDX version, one counts 214 inferences creating a much more complex HPDIC graph.