Content area
This dissertation addresses the persistent challenges faced by electrical engineering students in learning the fundamentals of electrical circuit analysis, a core subject regarded as demanding both theoretically and practically. To promote a more effective, autonomous, and engaging learning experience, this work builds on the U=RIsolve Academy educational platform — developed at the Instituto Superior de Engenharia do Porto — by restructuring its codebase to improve scalability and enable the future integration of additional learning modules with greater ease. One of the modules to be integrated is based on artificial intelligence, for which a dataset was created, bringing together materials used by first-year students in the Electrical and Computer Engineering programme at Instituto Superior de Engenharia do Porto, in the Fundamentals of Eletrotechnical Engineering and Circuit Theory courses. The trained model, named HALO assistant, was then integrated into the platform. This natural language processing is based on a model with a reduced number of parameters. Its performance was tested on content related to the nodal voltage method and the mesh current method through a blind evaluation alongside baseline models and the latest model released by OpenAI. The study involved more than 150 students and 18 domain experts, and the ongoing analysis is contributing to a scientific article in preparation with preliminary findings indicating that the trained models provide responses of comparable quality — and in certain aspects superior — to the significantly larger model that requires substantially greater computational resources.