Utilização de LLM com técnica RAG para geração de artefatos de contratação pública

Abstract

The modernization of public bidding processes, for through digitalization and automation, is essential to increase efficiency, transparency and compliance with legal standards. THE Law 14,133/2021 introduced regulations that require a more structured and detailed approach to artifact preparation of bidding, such as Preliminary Technical Studies (ETP) and Terms Reference (TR). This project aims to use information updated and accurate in the preparation of these artifacts, applying a Generative Artificial Intelligence (AI) model, based on Large Language Models (LLMs) and the Retrieval Augmented technique Generation (RAG), to overcome challenges in generating artifacts of public contracts. The results obtained demonstrated that the application of the proposed approach results in documents with better clarity, coherence and precision in comparison with texts generated only by LLMs with documents attachments. Assessments made by employees responsible for preparation of documents indicated an average final grade of 56/60 for RAG+LLM versus 45/60 for LLM. This work confirms that intelligent automation can increase efficiency, accuracy and compliance of bidding processes, benefiting both public institutions and society.