Learning Outcome Generation using LLM: Design and Validation

Abstract

This article explores the use of artificial intelligence to automate the generation of Learning Outcomes (LO) in higher education contexts. The proposal combines a Large Language Model (LLM) with a Retrieval-Augmented Generation (RAG) architecture, aiming to improve the accuracy, coherence, and pedagogical relevance of the generated texts. To achieve this, disciplinary document corpus and a database of LO previously validated by the educational community were integrated and used as contextual sources during the automatic generation process. The proposed architecture was implemented, and various experimental scenarios were analyzed using a single course, modifying input configurations such as prompt structure and model temperature. The results show that the system is capable of generating structurally correct LO, aligned with curricular parameters. As future work, the incorporation of automated mechanisms to assess pedagogical quality is proposed, along with extending the model to support the generation of other relevant educational artifacts.