The availability of face-to-face attendance at the School’s Administrative Services for Students is limited to one schedule, which may prevent the timely clarification of students' questions, causing a decrease in their level of satisfaction. To solve this problem, a conversational agent was designed, consisting of a Portuguese language interpretation module using natural language processing and machine learning techniques. To keep the system abstracted from any technical dependency, a web service that manages the agent's knowledge base was developed. In the evaluation of the solution, the performance of several learning models was compared, and the results emphasize the superiority of BERT language model of Google, combined with the DIET classifier, obtaining a F1-Score of 0.965. The system was implemented through a prototype and, for a total of 256 questions, around 70% of correct responses were obtained, with a positive average satisfaction rating of 4.20 on a 0-5 scale.