Artificial Intelligence · NLP · LLM Safety
M.Sc. student in Artificial Intelligence and NLP researcher at Universidade Federal de Goiás / CEIA.
I am an M.Sc. student in Artificial Intelligence at Universidade Federal de Goiás and an NLP researcher at the Center of Excellence in Artificial Intelligence (CEIA). My research focuses on language technologies for Portuguese speakers, with interests in LLM safety and alignment, minority-targeted toxicity, clinical AI, legal NLP, and practical dataset construction for specialized domains.
Across academic and R&D projects, I work on building and evaluating language models and datasets for socially grounded applications, including healthcare, law, energy, emotion recognition, and digital human interaction.
In Findings of the 64th Annual Meeting of the Association for Computational Linguistics (ACL 2026)
Bilingual benchmark showing that LLM safety and alignment vary by targeted minority group.
In Proceedings of the 15th edition of the Language Resources and Evaluation Conference (LREC 2026)
Synthetic dataset covering nine minority groups for Portuguese hate speech detection.
In Proceedings of the 15th edition of the Language Resources and Evaluation Conference (LREC 2026)
380k+ real patient-doctor QA pairs for Brazilian Portuguese clinical AI.
In Proceedings of the IEEE Conference on Virtual Reality and 3D User Interfaces Abstracts and Workshops (VRW 2026)
Studies how personality traits in digital human avatars affect engagement and communication in VR medical training.
In Findings of the Conference on Empirical Methods in Natural Language Processing (EMNLP 2025)
Survey of methods for personality in LLMs.
In Findings of the Conference on Empirical Methods in Natural Language Processing (EMNLP 2025)
Proxy LLM layer to intercept jailbreaks and prompt-leakage attacks before they reach the core model.
In Proceedings of the 19th International Workshop on Semantic Evaluation (SemEval 2025)
Studies data quality, synthetic augmentation, and intensity labels for Portuguese emotion recognition.
In Proceedings of the Brazilian Symposium on Computing Applied to Health (SBCAS 2025)
RAG chatbot for evidence-based reproductive and general health information for Portuguese-speaking women.
Universidade Federal de Goiás
Universidade Federal de Goiás
GPA/grade: 9.3/10
Research projects at the Center of Excellence in Artificial Intelligence (CEIA), developed with external partners.
Conversational Platform
Part of the application team building a conversational platform that helps Brazilian entrepreneurs understand and manage their businesses through natural-language questions, such as inventory and operational queries. I am responsible for data ingestion pipelines in a Google Cloud Platform environment and for conversational agent pipelines.
Energy-sector
I work on AI safety for LLMs in the electric power sector, focusing on security evaluation in Portuguese and English scenarios, harmful-content classification, and mitigation strategies to prevent undesirable model generations.
Legal NLP
Long-running R&D project on legal document understanding, retrieval, question answering, and domain-specialized generation for the Brazilian legal domain.
Cycle 3 · Jan 2026–present: Lead the text generation team, working with long context, information extraction, domain-specialized legal text generation, and question answering.
Cycle 2 · Jan 2024–Dec 2025: Tested and obtained results with retrieval-augmented generation, using classified documents together with retrieved and crawled jurisprudence and other legal documents; guided two junior researchers.
Cycle 1 · Jan 2023–Dec 2023: Worked on data curation and fine-tuning BERT-like models for legal document classification in Portuguese.
Legal NLP
Worked on a United Nations Development Programme (UNDP)-funded initiative with the National Justice Council (CNJ) to improve AI support for clustering repetitive legal claims and precedents. I used similarity algorithms, including Levenshtein distance, to disambiguate named entities and improve clustering quality.