Writing a thesis involves complying with certain requirements and rules established by institutional guides of universities. Students have guidelines to follow when developing their first draft, being insufficient to obtain a good document. This study seeks to help the students improve their first writings, based on natural language processing techniques. We focus primarily on the conclusions section of a thesis, a central element when finishing the research. In this paper, we present a conclusion analyzer that includes three models: goal connectedness, judgment and speculation. Such subsystems try to take care of main expected features in conclusions, specifically the connectedness with the general objective, the evidence of value judgments, and the presence of future work as a result of the student reflection. The study provides initial models, internal exploration of conclusions, and evaluations of our approach. We found in the three features evaluated that graduate level student’s texts outperformed those of undergraduate level. This behavior provides evidence that students with more practice writing a scientific paper or thesis (graduate level), possess better writing skills.