Public servants and auditors must often identify precise legal evidence across large collections of laws, decisions, audit reports, and technical documents. However, institutional search systems typically rely on rigid metadata or exact keyword matching, making it difficult to answer natural-language questions whose evidence is distributed across heterogeneous sources. This article presents a Retrieval-Augmented Generation (RAG) framework for legal corpora that combines hybrid retrieval, keyword extraction, metadata-driven filtering, and multi-hop retrieval for legal-administrative question answering. The study follows a staged design. Stage I establishes a keyword-augmented hybrid RAG baseline that integrates dense and lexical retrieval with automatic keyword extraction. Stage II extends this baseline with query decomposition, LLM-guided metadata extraction for dynamic filtering, and domain-specific cross-encoder reranking. We evaluate the framework on two corpora: a test set with more than 7,000 legislative and administrative documents and a validation corpus with more than 333,000 PDFs and 10 terabytes of data across five institutional knowledge bases: legislation, decisions, audit reports, prosecutor opinions, and portal files. Results show that keyword-guided hybrid retrieval improves evidence coverage and ranking quality, while structured multihop control further improves retrieval coverage and answer correctness in heterogeneous, high-stakes legal corpora.
Eniedson Fabiano Pereira da Silva, C. de Souza Baptista, A. L. F. Alves et al.· International Journal of Sem...· 0 citations
This paper presents an IoT Analytics architecture for real-time environmental monitoring, designed to overcome the limitations of solutions that provide limited support for operational and historical data analysis. The proposal integrates continuous collection, stream processing, temporal analytical storage, and interactive visualization via dashboards within a microservices framework. An end-to-end pipeline was implemented using open-source tools, and an evaluation was conducted in a real-world scenario in the state of Acre, Brazil, in comparison with the platform currently used for air quality monitoring. In the usability evaluation, the proposed solution achieved mean scores between 6.20 and 6.65 (on a Likert scale from 1 to 7), while the reference solution ranged between 2.15 and 2.50. The results indicate consistent gains in real-time indicator retrieval, historical time-series exploration, and the execution of analytical tasks.
P. M. Alves, Cláudio de Souza Baptista, A. L. F. Alves· Anais do XVII Workshop de Co...· 0 citations
The exponential growth in medical imaging volumes necessitates scalable, reliable diagnostic support systems capable of augmenting clinical workflows. This article presents a systematic quantitative evaluation of state-of-the-art Multimodal Large Language Models (MLLMs) for radiology Visual Question Answering (VQA), a task requiring integrated visual perception and clinical reasoning. We benchmark five leading models — GPT5-Nano, Gemini 3 Flash, Qwen3-VL-8B, LLaVA Next, and Llama 3.2 Vision — on the VQA-RAD dataset under a rigorous zero-shot protocol with standardized prompts and comprehensive precision–recall–F1 evaluation. Our empirical analysis reveals that Gemini 3 Flash achieves superior balanced performance (F1 = 0.78, Accuracy = 0.78, Recall = 0.83), while Qwen3-VL-8B attains the highest precision (0.78) while also maintaining competitive recall. These outcomes demonstrate that general-purpose MLLMs can perform competitively with specialized medical models in tasks such as modality and organ recognition, but still struggle with abnormality detection and complex clinical reasoning. The findings reinforce that MLLMs currently serve best as assistive decisionsupport tools rather than autonomous diagnostic agents, and highlight the potential of retrieval-augmented and context-aware strategies for improving clinical reliability and interpretability.
Cristovão Pessoa Cândido, Matheus Alves de Oliveira Lima, C. de Souza Baptista et al.· International Journal of Sem...· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
