Development of a virtual assistant as academic support for the Computer Science teaching degree program using customized generative AI and RAG

Abstract

This article presents the development of a virtual assistant aimed at the bachelor's degree in Computing at the Federal Institute of Tocantins – Araguatins Campus, with the purpose of automating responses to frequently asked questions from the academic community. The initiative seeks to provide a faster and more accessible support channel, especially in contexts with limited human resources for continuous service. The solution uses Artificial Intelligence techniques, with large language models (LLM) for the contextualized generation of automatic responses. The system was integrated with WhatsApp, allowing users to ask questions and receive real-time responses based on pre-organized and validated institutional course documents. Methodologically, the approach adopted was retrieval-augmented generation (RAG), which combines the retrieval of relevant documents with the semantic representation of texts through embeddings. User queries are converted into vectors and compared to a vectorized database, enabling the retrieval of the most relevant content. The documents are organized and maintained on Google Drive, which facilitates their curation and continuous updating. The virtual assistant was designed to answer frequently asked questions (FAQ) from the academic community, covering topics ranging from basic information, such as curriculum structure, workload, and class schedules, to content extracted from more complex institutional documents, such as bylaws, regulations for final projects and internships, meeting minutes, public notices, and other internal rules of the course.