Cidadão.Ai: Sistema Multi-Agente De Inteligência Artificial (IA) Para Democratização Do Acesso A Dados De Transparência Governamental Brasileira

This undergraduate thesis presents Cidadão.AI, a multi-agent system based on artificial intelligence to democratize access to government transparency data in Brazil through conversational interfaces. Using Design Science Research (DSR), the project combines natural language processing (NLP), multi-agent architecture, and LLMs to transform colloquial queries into complex analyses of public spending and administrative irregularities. The system integrates 17 specialized agents, each named after a Brazilian historical figure, responsible for domains such as anomaly detection and report generation. The architecture employs FastAPI (back-end), Next.js (front-end), and RAG for LLM integration. The implementation demonstrated technical feasibility through fundamental components including multi-agent structure, documented API, and cloud deployment. The main contributions are: (i) extensible and culturally contextualized architecture; (ii) framework for agents specialized in public data; (iii) DSR model applied to civic technology; (iv) collaborative open-source codebase. Results indicate that the multi-agent approach with advanced NLP can reduce barriers to public information access, strengthening social control and citizen participation.