Inteligência artificial no controle preventivo das contratações públicas: desenvolvimento e validação do agente edite para análise de editais

Abstract

This research aims to develop and validate an artificial intelligence agent named EDITE, designed for the preventive analysis of public procurement notices, with the capability to identify documental inconsistencies, legal risks, and clauses potentially restrictive to competition during the preparatory phase of public contracting. The study is situated within the field of Public Administration and digital transformation, engaging with theoretical approaches related to artificial intelligence applied to the public sector, governance, and procurement control, particularly under the framework of Law No. 14.133/2021. The methodology adopts a qualitative, exploratory, and applied approach, based on the documentary analysis of 300 procurement notices collected from the Brazilian National Public Procurement Portal, distributed across federal, state, and municipal levels. From this analysis, recurring patterns of inconsistencies were identified through expert evaluation, subsequently systematized and converted into analytical parameters that supported the development of the EDITE agent, structured through operational instructions that guide its performance in reading, interpreting, and comparatively analyzing procurement documents. The validation of the agent was conducted through its application to notices distinct from the development dataset and to real procurement processes within the institutional context of the research, with results compared to expert analysis. The findings demonstrate a high level of convergence between automated and human analyses, indicating the system’s ability to reproduce qualified analytical patterns, as well as to systematically identify relevant inconsistencies, such as documental discrepancies, weaknesses in object definition, and potential risks to competitiveness. The main contribution of the study lies in proposing an artificial intelligence model oriented toward preventive control, in contrast to prevailing post-audit approaches, with its originality evidenced by the conversion of empirical knowledge into operational guidelines applied through artificial intelligence and by the validation of the agent in real-world contexts. In terms of practical contributions, the use of EDITE showed potential to improve the quality of procurement notices, reduce failures in the preparatory phase, support decision-making by public agents, and enhance administrative efficiency, including indications of reduced processing time in procurement procedures, demonstrating its applicability across different contexts within Public Administration.