Name: OTAVIO LUBE DOS SANTOS
Publication date: 23/06/2026
Examining board:
Name |
Role |
|---|---|
| ALBERTO NOGUEIRA DE CASTRO JUNIOR | Examinador Externo |
| CAMILA ZACCHE DE AGUIAR | Coorientador |
| CREDINÉ SILVA DE MENEZES | Examinador Externo |
| DAVIDSON CURY | Presidente |
| MONALESSA PERINI BARCELLOS | Examinador Interno |
Pages
Summary: Learning Ecosystems (LEs) are complex environments where multiple actors, resources,
and technologies come together to support lifelong learning. Multi-Agent Systems (MAS)
enhanced by Generative Artificial Intelligence and Large Language Models (LLMs) open up
possibilities for creating intelligent and adaptive components within these ecosystems. Building
such systems, however, runs into conceptual, methodological, technological, and pedagogical
challenges. There is still a lack of approaches that systematically guide this development
in the context of LEs. This thesis addresses this gap by proposing LE-MASF (Learning
Ecosystems Multi-Agent Systems Framework), a conceptual and methodological framework
built upon the Design Science Research (DSR) methodology. LE-MASF combines two elements:
a six-layer reference architecture, which organizes the components of an educational MAS,
and a five-phase construction process (LE Analysis, Organizational Design, Agent Design,
Implementation and Integration, and Evaluation and Refinement). Its semantic base is LE-
MASO (Learning Ecosystems Multi-Agent Systems Ontology), a reference ontology grounded
in the Unified Foundational Ontology (UFO), modeled in OntoUML, and implemented in
OWL 2 DL. With 36 concepts organized into four sub-ontologies, LE-MASO provides a
shared vocabulary that preserves coherence between the MAS and the LE in which it operates.
The framework’s applicability was demonstrated through the development of LE-MASB
(Learning Ecosystems Multi-Agent System Builder), a web platform that operationalizes
LE-MASF’s principles and enables the assisted construction of educational MAS. Validation
followed three progressive strategies. The first was the direct application of LE-MASF in the
Intelligent Tutoring Systems course (PPGI/UFES), conducted by the author. The second was
the implementation of the framework’s principles in LE-MASB. The third was the independent
application by two educators, in contexts of neurodiversity and high-ability/giftedness. The
results indicate that LE-MASF supports the construction of MAS in Learning Ecosystems,
bringing together an ontological base, pedagogical grounding, and the use of generative agents.
