Logo do repositório

Percorrer por autor "Telmo Sampaio"

A mostrar 1 - 2 de 2
  • 3D-Chatbot: A Intelligent Conversational Agent Integrating RAG, NLP, And VR For Commercial Stores
    Publication . Telmo Sampaio; Oliveira, Pedro Filipe; Matos, Paulo
    This paper presents a comprehensive 3D chatbot system that combines advanced AI capabilities with practical commercial applications in immersive virtual reality environments. The system integrates natural language processing, retrieval-augmented generation (RAG) architectures, and real-time speech processing to create seamless customer inquiry experiences. The implementation combines four main components: an intelligent data acquisition pipeline that transforms over 1,000 unstructured web documents into AI-ready knowledge bases; an advanced RAG system utilizing OpenAI's text-embedding-3-small model and Pinecone vector database for semantic search; multimodal speech processing with real-time speech-to-text conversion and lip-sync animation; and a Unity-based VR environment with professional avatar interactions. The results demonstrate exceptional performance with a 94.6% success rate across 65 test interactions, achieving response times under 100 milliseconds for knowledge retrieval and completing full voice response cycles in less than 3 seconds. The system maintains conversation context through a six-step interaction flow while accessing external knowledge sources, with 96.9% contextually appropriate responses and 92.3% conversation continuity in multi-turn dialogues. Performance optimizations ensure responsive operation under variable load conditions, demonstrating the viability of intelligent virtual assistants in three-dimensional commercial spaces.
    2025Comunicação em conferência Acesso restrito Ver mais
  • A Chatbot to Enhance the Portuguese National Tourism Website
    Publication . Telmo Sampaio; Oliveira, Pedro Filipe; Matos, Paulo
    The digital transformation of tourism in Portugal has led to the integration of AI-driven solutions to improve visitor engagement. This paper presents the development and implementation of a chatbot for the Portuguese national tourism website, VisitPortugal. The chatbot utilizes natural language processing to provide real-time, personalized assistance across multiple languages, including European languages (Portuguese, English, French, Spanish) and East Asian languages (Traditional Chinese), demonstrating robust cross-linguistic semantic understanding. The architecture incorporates a web crawler to collect and process tourism-related data, a vector database for efficient content retrieval, and GPT-4o Mini for generating contextually relevant responses. By filling gaps in tourism services, the chatbot improves the user experience, increases accessibility, and reinforces Portugal's position as a technologically advanced tourist destination. The results demonstrate high accuracy in the responses, reinforcing the effectiveness of the chatbot in facilitating seamless travel planning.
    2025Comunicação em conferência Acesso restrito Ver mais