The Brazilian Supreme Court (Supremo Tribunal Federal - STF), the highest instance of the Brazilian judiciary system, produces an immense amount of data, usually organized in text form, through decisions, petitions, preliminary injunction, recourse/appeal and other legal documents. In this context, a supervised learning tool for classifying legal documents into the 17 objectives (SDGs) of the UN 2030 Agenda for Sustainable Development can be of great use to the court, given that this task is performed manually by a large group of court staff and the adoption of the UN 2030 Agenda is one of the main goals of the court currently. The general objective of this project, called RAFA 2030, in this way, is to generate value for the court through the construction of classification systems based on Natural Language Processing - NLP in order to allow STF officials (employees) responsible for the initial analysis of legal documents they just need to validate classifier responses in their daily workflows. Currently, the main methods used in this project consist of graphical tools for NLP (Co-ocurrence graphs), neural network algorithms, keyword counting and similarity analysis, as well as other tools available in R (main) and Python ​​(only Keras, Tensorflow and Spark NLP backend) languages. Initial results suggest immense potential for applications of NLP and deep learning to classify legal documents into UN 2030 Agenda themes.