talks

Bringing modern features to classic multi-agent systems

October 21, 2026 · 13:30 – 15:30 · Pantanal Auditorium

About the talk

Multi-agent systems and MAS frameworks have been around for several decades. However, the advent of agents based on Large Language Models (LLMs) gives rise to a crossroads moment in the history of MAS, when LLMs and classic agents can work together to leverage the advantages of each approach. LLMs are currently changing how agents work, behave, and interact. But the integration between classic and LLM-based agents, together with updates to the mechanics and tools offered by MAS frameworks, can bring great strengths and alleviate several weaknesses currently existing in both branches of multi-agent systems. In this context, classic MAS frameworks need an update. In this talk I will touch on several advances that we have done in our team, moving towards modern architecture and features for MAS frameworks.

Speaker

Andrei Olaru

University Politehnica of Bucharest, Romania

Andrei Olaru is Associate Professor at the National University of Science and Technology POLITEHNICA Bucharest — previously University Politehnica of Bucharest, where he serves in the Computer Science Department within the Faculty of Automatic Control and Computers. His academic activity focuses on artificial intelligence, with particular interests in intelligent agents, multi-agent systems, context-aware computing, and machine learning. Olaru obtained his Ph.D. in 2011 through a double-diploma doctoral program between Politehnica Bucharest and Université Pierre et Marie Curie (Paris 6, now Sorbonne University). His doctoral research addressed the integration of context-awareness in multi-agent systems for ambient intelligence applications. He is the co-founder and coordinator of the Master’s program in Financial Computing at Politehnica Bucharest, developed since 2017 in collaboration with industry partners. His research work includes participation in national and European projects related to artificial intelligence, distributed systems, and trustworthy AI, being member of many labs and actively contributing to research initiatives addressing both theoretical and applied aspects of intelligent systems. Olaru is also a founding member of the Romanian Association for Artificial Intelligence (ARIA).

Beyond the Model: Architecting Agentic AI Systems

October 22, 2026 · 09:30 – 10:30 · Vila Bela Room

About the talk

Large Language Models (LLMs) are evolving from conversational interfaces into agentic systems that reason, use tools, maintain memory, specialize, and coordinate toward goals. This shift suggests that AI capability is shaped not only by model size, but also by how systems are architected. In this keynote, I will discuss the principles behind agentic AI and argue for the importance of building effective systems on top of smaller, open LLMs, reducing dependence on a few proprietary developers and/or providers while improving technological sovereignty, cost efficiency, transparency, and accountability. As a case study, I will present ARARA, a modular multi-agent framework for conversational recommender systems, and show how explicit specialization, memory, and governed coordination can improve robustness and effectiveness. The broader lesson is that the future of AI may depend as much on how we architect and harness intelligence as on how large we make LLM models.

Speaker

Marcelo da Silva Reis

Institute of Computing, University of Campinas, Brazil

AAMAS Silver Jubilee: Where did we come from and where are we going?

October 19, 2026 · 17:00 – 18:00 · Vila Bela Room

About the talk

In 2026, the International Conference on Autonomous Agents and Multi-Agent Systems (AAMAS) celebrates its 25th edition. Since its first edition in Bologna in 2002, the field has made remarkable progress, generating new models, paradigms, and distinct scientific perspectives, inspired by areas as diverse as computer science and engineering, social sciences, psychology, logics, economics, and management, among others. This presentation will provide an overview of the field's history, origins, main research topics, and its place within the current landscape of integrating Artificial Intelligence in the corporate and social world.

Speaker

Jaime Simão Sichman

Polytechnic School, University of São Paulo, Brazil

Talk BRACIS

Challenges for the next generation of intelligent systems in the financial market

October 20, 2026 · 14:30 – 15:30 · Cerrado Auditorium

About the talk

Recent advances in Artificial Intelligence have opened new possibilities for building increasingly robust solutions, and have brought into focus scientific and technological challenges that remain open. In this talk, Eduardo R. Hruschka will discuss how these challenges surface in the development of intelligent systems for the financial market, covering topics such as robustness, hyper-personalisation, governance and generative AI. Drawing on hands-on experience in one of the sectors that invests most heavily in AI worldwide, the presentation will explore real-world applications — including conversational assistants, financial experiences embedded in digital channels, and intelligent systems for recommendation and decision support — and will also discuss research opportunities at the frontier between academia and industry.

Speaker

Eduardo R. Hruschka

Itau Bank

Talk BRACIS

EvidencIAJud: Inteligência Artificial Aplicada ao Apoio Decisório na Judicialização da Saúde

October 19, 2026 · 13:30 – 14:30 · Cerrado Auditorium

About the talk

A judicialização da saúde no Brasil é um fenômeno crescente no qual os tribunais são acionados para garantir demandas por medicamentos e/ou tratamentos, geralmente de alto custo, não cobertos pelo SUS (Sistema Único de Saúde) ou pela saúde suplementar. Peritos médicos elaboram pareceres técnicos específicos para cada caso contendo análises embasadas em evidências para eficácia, benefícios esperados e custos para fundamentar decisões judiciais que equilibrem as necessidades individuais do paciente e as restrições de recursos da sociedade. O EvidencIAJud é uma ferramenta de software assistida por IA que utiliza um fluxo de trabalho estruturado para automatizar a geração de relatórios preliminares, dando apoio a peritos médicos e juízes. O EvidencIAJud integra conhecimentos confiáveis de diversas fontes oficiais sobre os detalhes clínicos, regulatórios e de precificação de tratamentos autorizados para gerar análises de custo-benefício de alta qualidade para os relatórios. Atualmente, a ferramenta está em fase de avaliação por um grupo selecionado de juízes e médicos especialistas.

Speaker

Bruno Padilha

USP

Talk BRACIS

Gestão de Memória e Contexto em Agentes Conversacionais de IA

October 19, 2026 · 14:30 – 15:30 · Cerrado Auditorium

About the talk

Os avanços recentes em modelos de linguagem permitiram criar agentes conversacionais de IA capazes de fazer tarefas longas, com várias etapas e várias ferramentas. Ao mesmo tempo, mostraram que decidir o que o agente lembra, esquece e coloca em seu contexto ainda é um problema científico e tecnológico em aberto. Nesta palestra, Marlesson Santana discute como esse problema aparece no desenvolvimento de agentes conversacionais em produção. Os temas incluem memória de curto e de longo prazo, a janela de contexto como recurso finito, compressão e sumarização de histórico, recuperação seletiva de informação e degradação da qualidade em conversas longas e com vários assuntos. A palestra parte da experiência prática com o Jus IA, o assistente de inteligência artificial jurídica do Jusbrasil. Ela mostra aplicações reais, como o uso de ferramentas de busca e análise de documentos, a continuidade de conversas ao longo do tempo e a avaliação sistemática da capacidade de memória dos agentes. Também discute oportunidades de pesquisa na fronteira entre academia e indústria.

Speaker

Marlesson Santana

Jusbrasil

Staff Machine Learning Engineer — Jusbrasil.

Talk BRACIS

Mecanismos de incentivo, fontes de financiamento e arranjos de parceria entre ICTs públicas, ICTs privadas e empresas

October 22, 2026 · 14:30 – 15:30 · Cerrado Auditorium

About the talk

A palestra abordará os mecanismos de incentivo à inovação, as principais fontes de financiamento para projetos de pesquisa, desenvolvimento e inovação (PD&I) e as oportunidades de captação de recursos. Serão apresentados modelos de parceria entre ICTs públicas, ICTs privadas e empresas, com foco na colaboração entre academia e setor produtivo. A apresentação será conduzida por Gabriella Nóbrega, Analista Sênior de PD&I do Instituto Hardware BR.

Speaker

Gabriella Nóbrega

Instituto Hardware BR

Gabriella Nóbrega, Analista Sênior de PD&I do Instituto Hardware BR, é Gestora Pública por formação, especialista em Gestão e Docência do Ensino Superior e Mestranda em Ciência da Informação pela Universidade de Brasília (UnB).

Talk BRACIS

Observatório da Agenda Nacional

October 22, 2026 · 08:30 – 09:30 · Cerrado Auditorium

About the talk

A palestra apresenta um mapeamento da formação em nível de graduação nos temas estratégicos dos estados brasileiros, com destaque para o estudo sobre a formação em Inteligência Artificial no cenário nacional. A análise contempla a distribuição e as características dos cursos e competências relacionadas à IA no Brasil, evidenciando os desafios e as oportunidades para alinhar a formação acadêmica às demandas estratégicas do país.