Repository logo

Browsing by Issue Date, starting with "2025-11-14"

Filter results by year or month
Now showing 1 - 2 of 2
  • Desenvolvimento Web Full-Stack na Empresa Tetrapi SA
    Publication . Santos, Dário Vítor Freitas; Correia, João António Câmara; Sousa, Maria Paula Prata de
    Este documento descreve o estágio realizado para a unidade curricular Dissertação ou Estágio no segundo ciclo de Engenharia Informática da Universidade da Beira Interior. Este estágio focou-se no desenvolvimento de uma plataforma web na empresa Tetrapi SA com recurso à framework Symfony. O estágio teve como objetivo principal o desenho e desenvolvimento de uma plataforma web para a Subdireção Regional dos Transportes Terrestres, e teve como objetivos adicionais a orientação de outras equipas de desenvolvimento, a orientação de um estágio curricular e a manutenção de uma plataforma web em fase de validação pelo cliente. Com o intuito de apresentar o máximo de informação possível, o presente relatório está dividido em múltiplos capítulos que abordam a empresa e o plano de estágio; o estado da arte no que consta ao desenvolvimento de plataformas web na framework Symfony; e o desenvolvimento propriamente dito da plataforma Sistema de Gestão de Contra Ordenações (SGCO).
    2025-11-14Master thesis Open access Show more
  • Conversational Agent for Mental Health Support
    Publication . Silva, Luís Gonçalo Aguilar; Pais, Sebastião Augusto Rodrigues Figueiredo
    Mental health is one of the greatest global challenges today, marked by an increase in cases of anxiety and depression and difficulties in accessing specialised support, high costs and stigma. In this context, digital solutions that expand the scope of care are a promising approach. This dissertation presents the design and implementation of a conversational agent for mental health support. The main model used was Gemma 3 4B, applied with prompt engineering. LoRA tuning was explored but not adopted due to export constraints. This LLM is complemented by models for Hate Speech Detection, Emotion Detection, Sentiment Detection, and Sarcasm Detection. The BERT classifiers (sentiment, emotions, sarcasm, and hate) generate signals that condition the tone and dialogue strategy, with these responses being integrated into the system prompt so that the interaction is dynamic. The architecture supports text data input and output and contains modules for periodic check-ins, detection of signs of risk of self-harm or despair with the provision of national suicide helpline contact details, and medication management through the sending of PDF prescriptions and scheduling of medication reminders. Experiments show that the proposed methodology produces more empathetic, confident, and useful responses than using only a language model in varied scenarios. The cosine semantic similarity was calculated to compare the agent with market applications and with experimental cuts without the analysis modules, and the scores indicated comparable content, while the agent’s responses were notably more validating, empathetic, safe, and practice-oriented. Added components such as regular check-ins with the user and medication scheduling to provide reminders indicate practical potential as a virtual therapist. Limitations include occasional errors in detecting whether a direct response or reflection is desired, loss of context due to short memory, and incomplete speech and avatar components.
    2025-11-14Master thesis Open access Show more