Overseeing Government with AI : Lessons learned from a Brazilian experience

This paper reviews a project carried out in the Brazilian Congress that, since 2020, monitors and selects relevant articles from the federal government gazette using Natural Language Processing and Machine Learning as part of an executive branch overseeing effort. The selected articles are also used to produce a daily bulletin that is open to the general public. The technical details are presented, and the successes and possible improvements are highlighted. Overall, the project is evaluated as a positive use of AI in government with minimal adverse side effects.