Prototipo de preguntas-respuestas basado en una arquitectura RAG. Dominio de aplicación: Turismo en el Ecuador

Abstract

Abstract: Technological advances have facilitated the creation, dissemination and global distribution of content almost immediately. Advances in large language models (LLM) have facilitated the development of natural language processing (NLP) applications such as Q&A systems. This work is oriented towards the development of a Q&A question-answer system based on a RAG architecture for the tourism application domain in Ecuador. The system component diagram includes elements such as user, query, emdeddings, extractive model. and generative and finally the response. The extractive model "sentence-transformers/all-mpnet-base-v2" searches for relevant information in the database created for the project, providing context to the generative models "meta-llama/Meta- Call-3-8B-Instruct" and "google/gemma-7b-it" to produce coherent answers. The results indicated a reasonable performance in information retrieval and a superiority of the "meta-llama/Meta-Llama-3-8B-Instruct" model in text prediction. The interface, developed with Streamlit, offers a friendly experience for users, allowing effective consultations about tourism in Ecuador. By providing clear and accurate information, the system has the potential to promote tourism in Ecuador, contributing to local economic development.