Explainability remains a major challenge in AI-powered retrieval systems, particularly when large language models generate answers without transparent reasoning. We introduce a structured grounded reasoning framework that operationalizes explainability as a compositional property derived from explicit semantic provenance and rule-guided inference. Our approach integrates knowledge graphs, SPARQL-based retrieval, and dynamic prompt generation to ensure that responses are grounded in verifiable structured data and accompanied by interpretable reasoning traces. We formally define explainability criteria and implement the framework within a recommendation scenario. Experimental results show that our method significantly improves interpretability, traceability, and user trust compared to standard retrieval-augmented generation baselines while preserving retrieval accuracy. These findings demonstrate that explainability in LLM-driven systems can be systematically engineered rather than heuristically approximated.

Retrieval-Augmented Generation (RAG) systems combine large language models with retrieval mechanisms to generate contextually relevant answers. However, while vector similarity effectively retrieves geometrically close documents, it often lacks the semantic depth required to fully address user queries. This paper investigates the gap between vector similarity and semantic relevance through mathematical formulations and real-world examples. We propose SemantriX, an explainable hybrid retrieval strategy that integrates metadata enrichment and cross-encoder-based reranking. The model is evaluated within a question-answering system applied to the domain of contract management. Experimental results show significant improvements in precision, recall, and F1-score. By aligning retrieval with semantic relevance, our approach enhances the performance and explainability of RAG systems in real-world decision-support scenarios.

Background Convergent evolution, the independent emergence of similar traits, is increasingly recognized as a pervasive force shaping molecular and metabolic diversity. A striking manifestation of convergence at the molecular level is represented by non-homologous isofunctional enzymes (NISE), distinct proteins with no detectable common ancestry that catalyze identical biochemical reactions. Despite their conceptual and practical relevance, NISE are often treated as exceptional cases, and no large-scale, systematically curated resource has been available to explore their distribution and properties across all domains of life. Data description Here we present a curated dataset of homologous and non-homologous isofunctional enzymes (HISE and NISE) derived from UniProtKB release 2025_01, encompassing both reviewed (Swiss-Prot) and unreviewed (TrEMBL) entries. Using Enzyme Commission (EC) numbers to define catalytic equivalence and SUPERFAMILY (SCOP structure superfamily) annotations to infer evolutionary relationships, we implemented a transparent and reproducible pipeline to classify enzymes into homologous and non-homologous functional groups. The dataset comprises over 200,000 Swiss-Prot and 27 million TrEMBL enzymes with complete EC and SUPERFAMILY annotations, organized by domain of life, enzyme class, and structural domain composition. Multiple output files, including presence/absence matrices, clustered enzyme groups, phyloprofiles, and full annotation tables, are provided to facilitate downstream evolutionary, functional, and comparative analyses. This resource offers a global view of molecular convergence and divergence in enzymatic functions, highlighting the widespread nature of NISE across taxa and enzyme classes. It provides a foundation for studying metabolic evolution, functional redundancy, drug target discovery, and the evolutionary constraints shaping biochemical solutions.

In human-AI interaction, users often begin with vague, incomplete, or imprecise queries that limit the system’s ability to provide meaningful support. While much attention has been given to generating accurate answers, the capacity of intelligent systems to guide users in formulating better questions remains underexplored. This paper proposes a methodology that supports users in the co-construction of knowledge through context-aware clarification questions. By combining semantic representations from knowledge graphs with the generative capabilities of large language models, the approach dynamically detects informational gaps and offers targeted prompts that refine user intent. This method fosters more natural, adaptive, and cognitively aligned interactions, improving user engagement, decision-making, and exploratory search. We discuss implications for designing intelligent systems that not only respond, but also proactively assist users in asking better questions.

Contract management in public institutions demands rigorous oversight of service execution and financial compliance, posing operational challenges for managers. This study introduces an LLM question-and-answer (Q A) system to retrieve structured and unstructured contract information. The solution integrates text-to-SQL extraction for data querying, prompt engineering for response standardization, and intelligent agents for context-aware answer generation. Implemented as a containerized Python application and combining generative AI with customized LLMs, the system overcomes the limitations of traditional contract management systems. The results demonstrate an effective architecture for deploying a Q A system to address one of the main challenges in understanding administrative contracts and their specifics.

Explainability is a key aspect of data retrieval systems, particularly when leveraging Large Language Models (LLMs) to build question answer systems. This study proposes a methodology to enhance explainability in retrieval processes, ensuring that results are not only accurate but also interpretable. By integrating Knowledge Graphs (KGs) with dynamic prompt engineering, we systematically guide LLMs to generate transparent and contextually justifications for their outputs, making retrieval decisions more comprehensible to users. To evaluate the effectiveness of this method, we implement it within a Recommender System and assess its impact on user trust and decision-making. Experimental results demonstrate that our method significantly improves the interpretability of retrieval outcomes while maintaining high retrieval performance—all without requiring model retraining. This work highlights the potential of combining structured knowledge representations with prompt engineering to bridge the gap between AI performance and user-centric explainability.

This article explores the utilization of the Hadoop ecosystem as a polyglot big data processing platform, focusing on the integration of diverse computation and storage technologies and their potential advantages in certain computational contexts. It delves into the potential of this ecosystem as a unified platform highlighting its architectural foundations and their complementary strengths in distributed storage, processing efficiency and real-time analytics. The article explores potential use cases within domains such as Smart Cities and Social Networks, illustrating how the platform's diverse components can be orchestrated in a polyglot manner and how these fields can benefit from the ecosystem's capabilities. Finally, the article concludes by showcasing alternatives for future research, including specialized architectural aspects of the ecosystem to advance the polyglot paradigm.

This paper presents an advanced question-answering (QA) system designed to support enterprise contract management by seamlessly integrating document and database information. Leveraging large language models (LLMs) and agent orchestration, our system delivers precise, context-aware responses to complex contract-related queries. We enhance retrieval accuracy through a pipeline incorporating Retrieval-Augmented Generation (RAG) and Text-to-SQL techniques, eliminating the need for LLM retraining. Through targeted Prompt Engineering, we refined the system's ability to extract and synthesize key contractual information, significantly improving response relevance and accuracy. Our evaluation demonstrates the system's potential to significantly reduce time-consuming tasks in contract workflows and provide actionable insights, marking a significant advancement in enterprise contract management systems.

Este artigo investiga o fenômeno do capitalismo de vigilância e suas implicações para a coleta e uso de dados pessoais na web. Através da análise empírica do tráfego de rede, demonstramos como as técnicas de rastreamento e direcionamento de dados, inicialmente desenvolvidas para fins comerciais, permeiam a experiência online dos usuários, revelando um ecossistema complexo onde informações pessoais são continuamente extraídas, compartilhadas e utilizadas para influenciar o comportamento dos usuários. Além disso, discutimos a preocupante possibilidade da aplicação dessas técnicas para a disseminação de notícias falsas e desinformação, representando uma ameaça direta à integridade democrática. Ao fornecer evidências empíricas, podemos contribuir para o desenvolvimento de políticas e ferramentas eficazes para proteger a integridade da informação e fortalecer a democracia na era digital.

We explore trustworthiness from a dual perspective, addressing both the content and the container (source). Through an illustrative example, we analyze the challenges and opportunities in assessing trustworthiness using the Reference Ontology of Trust (ONTrust) as well as the explanatory notion of Ontological Unpacking. The key findings reveal that establishing information trustworthiness necessitates a clear understanding of its underlying ontological commitments, followed by a systematic process of negotiating admissible truth-makers and deliberating on appropriate information proxies for these truth-makers. These results underscore the importance of explicitly defining the ontological assumptions behind information and systematically evaluating the proxies used, ensuring more robust and coordinated decision-making processes.

In exploratory search scenarios, users frequently turn to Knowledge Graphs (KGs) as rich sources of structured knowledge. However, user queries are often imprecise, incomplete, or ambiguous, and KGs themselves are inherently incomplete. Moreover, when KGs are constructed from crowd-sourced content or by integrating heterogeneous data sources of varying quality, the need for mechanisms to assess information reliability becomes critical. This work introduces Context-Augmented Knowledge Graphs (CoaKG), an enriched KG model designed to support context-aware exploration for decision-making and task learning. CoaKG incorporates both explicit contextual mappings and inference rules to derive implicit context, enabling more nuanced query answering. We also present the CoaKG Query Engine, a query processing framework to retrieve contextually enriched answers tailored to users' goals and context.

Este artigo propõe uma abordagem inovadora para aprimorar sistemas de recomendação por meio da integração de Grafos de Conhecimento (Knowledge Graphs – KGs) e Modelos de Linguagem de Grande Escala (Large Language Models – LLMs), utilizando triplas no formato Resource Description Framework (RDF) e engenharia de prompts. O sistema aproveita o conhecimento estruturado dos KGs para fundamentar as recomendações geradas pelos LLMs com explicações interpretáveis. Avaliamos o sistema em três cenários: geração de recomendações de produtos para usuários individuais, identificação de candidatos a descontos promocionais e seleção de combinações ideais de produtos. Os resultados demonstram que o método proposto gera sugestões contextualmente relevantes, acompanhadas de justificativas claras e compreensíveis para seres humanos. Também discutimos os principais desafios enfrentados durante o desenvolvimento e apresentamos estratégias eficazes para superá-los. De modo geral, nossos achados destacam o potencial dos recomendadores explicáveis baseados em LLMs para superar os sistemas tradicionais em termos de eficácia e transparência.

Transparency in public funds is essential for oversight and social control. However, some current monitoring solutions for Congressional Earmarks are incomplete. Official government sources are distributed across multiple endpoints, making data extraction challenging and creating user uncertainties. In addition, other initiatives are limited to specific types of congressional earmarks, failing to provide a comprehensive view of the process. This paper presents the development of a system capable of tracking the entire lifecycle of congressional earmarks, from proposal to payment. The solution was made with a relational database, allowing structured queries to identify authors, recipients, amounts, and congressional earmarks locations (municipal, state, and regional). Additionally, the system integrates Business Intelligence (BI) tools to generate interactive analytical views, facilitating data comprehension and supporting decision-making. With a clear and user-friendly interface, the platform improves transparency and is essential for researchers, journalists, and citizens interested in monitoring budget allocations.

mRNA translation involves complex concurrent processes in which many mRNA strands might be processed simultaneously, leading to parallel consumption of potentially scarce resources. Previous studies have shown that excess mRNA can harm protein synthesis efficiency. This work investigates the role of ribosomes as concurrency controllers in regulating translation dynamics. Using Petri net modeling, we demonstrate that ribosomes act as natural regulators that limit the detrimental effects of mRNA abundance by controlling the number of concurrent translation events. Our experimental results show ribosome availability is a protective mechanism against greedy consumption patterns that lead to incomplete protein synthesis. This finding provides insights into the regulatory mechanisms maintaining translation efficiency in cellular systems.

The accurate annotation of pseudogenes is a significant challenge in genomics, as their decaying sequences often fall into a "twilight zone" of similarity that confounds automated methods. This paper describes a robust, homology-based methodology designed to overcome this issue. The core of the approach is a shuffle-based statistical evaluation used to establish a custom, empirically-derived significance threshold. This allows for the confident discrimination of true, biologically significant sequence remnants from stochastic background noise, providing a reliable framework for annotating pseudogenes and unannotated coding sequences in large-scale genomic projects.

Apresentamos uma ferramenta que integra modelos de linguagem, documentos, bancos de dados relacionais e agentes autônomos para ampliar as capacidades tradicionais de consulta em linguagem natural. Diferente de soluções limitadas a respostas pontuais sobre dados, nossa proposta permite a realização de análises avançadas, geração de insights e previsões automatizadas, apoiando decisões complexas em contextos organizacionais. A ferramenta foi aplicada em um domínio real de gerenciamento de contratos de tecnologia, com foco em análise de riscos e controle orçamentário. O artigo apresenta uma avaliação qualitativa com usuários finais, evidenciando benefícios em termos de acessibilidade, agilidade analítica e suporte estratégico à tomada de decisão. O trabalho contribui com uma proposta prática e extensível para a construção de sistemas de apoio à decisão mediados por inteligência artificial.