Retomando explicaciones con un chatbot por IA

The expansion of generative artificial intelligence (GenAI) poses new challenges for higher education, particularly in Ibero-American contexts where empirical evidence remains scarce. This article presents an exploratory study analyzing the pedagogical use of an educational chatbot, based on Retrieval-Augmented Generation (RAG) architecture, developed and implemented at the Universidad Nacional de Río Negro (UNRN) in Argentine Patagonia. The system converts transcriptions of recorded lectures into a navigable semantic corpus, which allows 57 students to query the teaching discourse accumulated throughout the course. The study adopts a longitudinal mixed-methods design that combines quantitative analysis (of 355 messages recorded across 40 sessions) with qualitative analysis (of student interactions). The architecture is documented in a detailed and reproducible manner: n8n orchestration, Whisper transcription, OpenAI embeddings and generation, top-20 dense retrieval over a Pinecone vector database, and conversational memory in PostgreSQL. On this basis, the expected performance of its components and their sources of error are discussed. The conceptual framework articulates situated learning theory (Lave & Wenger, 1991) with the theory of conceptual fields (Vergnaud, 1990, 1996); this perspective emerged from the analysis of the data itself. Drawing on the empirical evidence and the architectural analysis, the category of augmented pedagogical memory is proposed as a conceptual tool for describing longitudinal RAG infrastructures closed to the pedagogical corpus. The results show a predominant use oriented toward conceptual clarification, procedural reorganization, and the recapitulation of situated content, with no expansion toward external frameworks.