https://repositorio.ifpe.edu.br/xmlui/handle/123456789/495 2026-09-08T22:50:28Z https://repositorio.ifpe.edu.br/xmlui/handle/123456789/2309 Sistema automatizado de criação de avaliações baseado nos descritores do SAEB utilizando LLM A avaliação educacional desempenha papel fundamental na mensuração da aprendizagem e no direcionamento de políticas públicas. No contexto brasileiro, o Sistema de Avalia- ção da Educação Básica (SAEB) constitui o principal instrumento de avaliação em larga escala, exigindo que as questões aplicadas estejam alinhadas a descritores específicos de competências e habilidades. Entretanto, a elaboração manual dessas avaliações demanda tempo, conhecimento técnico e cuidado metodológico por parte dos docentes. Este traba- lho apresenta o desenvolvimento de um sistema automatizado para criação de avaliações baseadas nos descritores do SAEB, utilizando Modelos de Linguagem de Grande Escala (LLMs). A solução proposta integra uma arquitetura em camadas composta por parame- trização, geração, avaliação e saída, permitindo a geração de questões textuais personali- zadas quanto ao nível de dificuldade e descritor selecionado. O sistema foi implementado com FastAPI no backend, Vue.js no frontend e integração com API de modelo de lin- guagem para geração automática de itens. A validação foi realizada por meio de testes piloto com quatro educadores do ensino médio, que avaliaram as questões geradas quanto à coerência, contextualização, clareza, nível de dificuldade e alinhamento ao descritor. Os resultados indicaram elevados índices de aceitação, com médias superiores a 4,0 em escala de 1 a 5, evidenciando a viabilidade da ferramenta como apoio ao planejamento pedagó- gico. Conclui-se que a aplicação de LLMs na geração automática de avaliações alinhadas ao SAEB apresenta potencial significativo para otimizar o trabalho docente, mantendo qualidade pedagógica e padronização avaliativa, desde que acompanhada de supervisão humana. 2026-03-09T00:00:00Z https://repositorio.ifpe.edu.br/xmlui/handle/123456789/2275 TreePilotics: processamento, métricas ecológicas visualização interativa para inventários florestais padronizados Standardized forest inventories are essential for monitoring and comparative studies of tropical ecosystems; however, their exploratory analysis often requiresprogramming skills, limiting accessibility for researchers and educators. This study presents TreePlotics, an R/Shiny dashboard that automates the ingestion, processing, and visualization of forest inventory data, focusing on portability, usability, and offline availability of its core functionalities. The system implements an ETL (Extract, Transform, Load) pipeline capable of validating input data, calculating ecological attributes, and providing aggregated metrics, taxonomic rankings, spatial maps, and customizable graphical visualizations. The application was packaged with R Portable and includes code protection mechanisms based on encryption. Tests using real data (618 records) showed that the ETL pipeline processing was completed in less than 3 seconds, whereas the portable version initialization required an average of 8.5 seconds. In a stress scenario with 147,000 records (330 plots), data processing was finished in 34.3 seconds, with reactive filter response times remaining under 3 seconds. A usability assessment with five participants with different academic backgrounds indicated high ease of use for most modules, faster data exploration compared with spreadsheets, and willingness to adopt the tool in future analyses. The results demonstrate that TreePlotics reduces technical barriers to forest inventory exploration, providing a portable, secure, and accessible tool for research and educational applications. 2026-07-21T00:00:00Z https://repositorio.ifpe.edu.br/xmlui/handle/123456789/2240 O impacto das metodologias cascata e Scrum na criação de aplicações utilizando Outsystems The advancement of low-code platforms, such as OutSystems, has transformed software development by reducing technical complexity and accelerating application delivery. In this context, the choice of development methodology plays a decisive role in project productivity and quality. This study provides a comparative analysis of the impacts of the Waterfall and Scrum methodologies in the creation of low-code applications, based on a literature review and market experiences observed between 2015 and 2026. The results indicate that the Waterfall model is more suitable for projects with a closed scope, regulatory environments, and contexts that require rigorous documentation, benefiting predictability and traceability. In contrast, Scrum enhances the low-code paradigm by fostering iterative cycles, incremental deliveries, and greater adaptability to change, thereby reducing rework and increasing customer satisfaction. It is concluded that the choice of methodology in the context of low-code and the OutSystems platform should consider the degree of requirement stability and organizational maturity, with Scrum being more compatible with dynamic scenarios and better supporting mature teams, while Waterfall proves more efficient in controlled and normative environments, facilitating stakeholder separation and serving as a viable option for less experienced teams. 2026-05-28T00:00:00Z https://repositorio.ifpe.edu.br/xmlui/handle/123456789/2212 Agente Inteligente baseado em LLM para orientação clinica em profilaxia pós- exposição a materiais biológicos Exposure to biological materials poses significant risks for HIV and other sexually transmitted infections, making the timely initiation and proper management of post-exposure prophylaxis (PEP) essential. Effective PEP, however, depends on accurate risk assessment, clear protocol guidance, and continuous follow-up—processes often limited by healthcare constraints and patient misunderstanding. These challenges may lead to delayed initiation or poor adherence to the 28-day regimen. To address this gap, this study presents a Large Language Model-based conversational Agent integrated with a Retrieval- Augmented Generation (RAG) framework to support patients undergoing PEP in Brazil. The Agent was assessed using a five-point Likert-scale questionnaire applied to eighteen medical specialists, who evaluated clinical accuracy, alignment with official protocols, empathy, and conversational coherence. The Agent’s responses were highly rated, with over 80% strong agreement across all evaluated dimensions. In the overall assessment, 77.8% of evaluators rated the system as high quality, while 16.7% considered it good with minor limitations, supporting its feasibility and potential applicability as a clinical guidance-support tool. 2026-03-09T00:00:00Z