Prototipo de consulta de documentos jurídicos, desde el enfoque positivista, basado en una arquitectura RAG

Abstract

Abstract: The present curricular integration work examines the implementation of large language models within the legal sector, along with the limitations that arise when they are used individually for generating concrete and well-argued responses. As the main contribution, the study proposes the implementation of a query prototype based on the Retrieval Augmented Generation (RAG) architecture, which incorporates efficient information retrieval techniques combined with generative models. Through this approach, improvements in the precision, coherence, and traceability of legal content can be achieved, thereby reducing the risk of drawing incorrect or premature conclusions. The work demonstrates the appropriate and responsible use of artificial intelligence, supported by norms and doctrines of Ecuadorian jurisprudence, which contributes to more transparent decision-making aligned with positivist ideology. The integrated components of the prototype include the user, the upload of judicial rulings in PDF format, embeddings, an extraction model, and generative models such as GPT-5, Gemini 2.5-Flash, and Ollama 3, aimed at providing reasoned responses through the identification of patterns present in a typical judicial ruling.