Chatbot multimodal de soporte a usuarios con LLMs y análisis de voztexto.

Abstract

Abstract: This curricular integration project aims to develop a multimodal chatbot for user support based on large-scale language models (LLMs), capable of interacting via text and voice to guide and address user queries. The project arose in response to the limitations of traditional systems, which struggle to provide personalized, contextualized, and coherent responses. For the successful development of this project, the Design Thinking methodology was implemented. This methodology allowed for the identification of user needs and, based on the information obtained, the design of a user-centered solution. Furthermore, a Retrieval-Augmented Generation architecture was integrated, combining the generative model with information extraction from documents, the vector database, and content obtained through web scraping that is, the system corpus. Speech-to-text transcription was also incorporated to enable multimodal interaction. The results show that the Chatbot generates accurate, coherent responses using verified information, improving the accessibility and efficiency of support, demonstrating the potential of Long-Term Learning Management Systems (LLMs) which are key tools for optimizing user support.