In the rapidly evolving domain of conversational agents, the integration of Large Language Models (LLMs) into Chatbot Development Platforms (CDPs) is a significant innovation. This study compares the efficacy of employing generic and fine-tuned GPT-3.5-turbo models for designing dialog flows, focusing on the intent and entity recognition crucial for dynamic conversational interactions. Two distinct approaches are introduced: a generic GPT-based system (G-GPT) leveraging the pre-trained model with complex prompts for intent and entity detection, and a fine-tuned GPT-based system (FT-GPT) employing customized models for enhanced specificity and efficiency. The evaluation encompassed the systems’ ability to accurately classify intents and recognize named entities, contrasting their adaptability, operational efficiency, and customization capabilities. The results revealed that, while the G-GPT system offers ease of deployment and versatility across various contexts, the FT-GPT system demonstrates superior precision, efficiency, and customization, although it requires initial training and dataset preparation. This research highlights the versatility of LLMs in enriching conversational features for talking assistants, from social robots to interactive chatbots. By tailoring these advanced models, the fluidity and responsiveness of conversational agents can be enhanced, making them more adaptable and effective in a variety of settings, from customer service to interactive learning environments.

Objectives: Socially assistive robots (SARs) have emerged as promising solutions for elderly care, offering companionship, cognitive stimulation, and therapeutic interventions. However, their evaluation presents unique challenges due to the multidisciplinary nature of these systems and difficulty assessing them in real situations. This study presents SHARA-WoZ, a multistakeholder framework for evaluating SARs through Wizard of Oz (WoZ) methods. Methods: The framework implements a modular web-based infrastructure comprising three components: robot simulation interface, central server, and desktop wizard control application, supporting autonomous, semi-autonomous, and full-manual operational modes. Evaluation employed a between-subjects study with healthcare professionals and technical experts ( n = 10 each) using the user experience questionnaire (UEQ), custom usability assessment of system components, and qualitative analysis of post-session interviews. Results: The framework achieved above-average ratings across all UEQ dimensions. Healthcare professionals showed consistently higher ratings than technical experts, with largest gaps in Efficiency ( Δ =0.46) and Dependability ( Δ =0.47), indicating successful alignment with clinical workflows and therapeutic requirements. Qualitative analysis yielded 55 improvement suggestions across Interface Design, New Functionalities, Technical Performance, Integration Features, and Robot Behavior categories. Conclusion: The findings demonstrate that modular, web-based WoZ infrastructures effectively bridge the gap between research prototypes and real-world deployment requirements. The study confirms the critical importance of multistakeholder evaluation approaches, where healthcare professionals provide clinical perspectives while technical experts contribute optimization insights, ensuring both technical excellence and clinical utility in assistive robotics development.

Los robots sociales asistenciales emergen como solución para el cuidado de personas mayores, pero su validación efectiva requiere metodologías que faciliten la participación de múltiples implicados técnicos y no técnicos. La técnica Wizard of Oz (WoZ) permite evaluar interacciones antes de su implementación autónoma completa, aunque las plataformas actuales carecen frecuentemente de infraestructuras estandarizadas y modulares. Este artículo presenta SHARA-WoZ v2.0.0, una evolución de una plataforma de evaluación para robots sociales asistenciales que aborda sistemáticamente las limitaciones identificadas en su versión anterior. Las contribuciones principales incluyen: (1) migración a una aplicación de escritorio nativa en PyQt6 que reduce la latencia de renderizado en un 42% y el uso de memoria en un 49%; (2) integración de GPT-4o-mini que elimina servicios de traducción intermedios y reduce la latencia conversacional total en un 55%; (3) sistema de generación de respuestas múltiples basadas en estados emocionales que reduce el tiempo de respuesta en modo semi-automático a 2-3 segundos; (4) rediseño modular de la interfaz que facilita la supervisión simultánea de múltiples canales de interacción; y (5) arquitectura orientada a servicios que garantiza escalabilidad y extensibilidad. Los resultados demuestran mejoras cuantificables en rendimiento, latencia y usabilidad, estableciendo una plataforma que equilibra sofisticación técnica con accesibilidad para implicados no técnicos en el desarrollo y evaluación de robots sociales para contextos geriátricos. Estos resultados pueden ser generalizables como modelo arquitectónico efectivo para la aplicación de técnicas de Mago de Oz para el contexto de los Robots Asistenciales y Sociales.