DECIDEAI: A CHAT-JT ASSISTANT BASED ON GENERATIVE ARTIFICIAL INTELLIGENCE AND RETRIEVAL-AUGMENTED GENERATION TO SUPPORT THE ANALYSIS OF APPEALS IN THE TRT21 (Regional Labor Court of the 21st Region).
Labor Law; Design Science Research; Algorithmic Governance; Human Supervision; Judicial Decision Support Systems
This dissertation is situated within the context of contemporary challenges facing the Brazilian Labor Courts related to procedural delays and the high cognitive load involved in the admissibility analysis of the Appeal to the Superior Labor Court. In this context, the study aims to develop and demonstrate a technological artifact for institutional support, called DecideAI, conceived as a specialized assistant for Chat-JT, the Labor Courts' Artificial Intelligence platform, based on Generative Artificial Intelligence and the Retrieval-Augmented Generation architecture. The research adopts Design Science Research as its method, structured from the identification of the problem, the theoretical-normative foundation, the design and development of the artifact, as well as its demonstration and initial evaluation of its suitability for the purpose. DecideAI was designed to support, under continuous human supervision, activities of screening, verification of formal requirements, document organization, and reasoned suggestion of drafts, in accordance with the applicable normative framework, especially the guidelines of the National Council of Justice and the Superior Council of Labor Justice. As a result, an assistive architecture is presented, governed by mechanisms of explainability, traceability, and institutional control, capable of contributing to procedural efficiency without replacing human judgment. It is concluded that the proposed solution demonstrates technical and institutional viability as a tool to support appeals analysis, although it is limited to an initial validation, with further empirical evaluations planned.