From linear syllabi to dynamic knowledge graphs: an llm-powered multi-agent system for co-creative curriculum design and proactive analysis

Sánchez Sánchez, Nelson

From linear syllabi to dynamic knowledge graphs: an llm-powered multi-agent system for co-creative curriculum design and proactive analysis

dc.contributor.advisor	Galpin, Ixent
dc.creator	Sánchez Sánchez, Nelson
dc.date.accessioned	2026-02-20T21:48:30Z
dc.date.created	2026-02-18
dc.description.abstract	La arquitectura curricular contemporánea se ve afectada por una "opacidad estructural", la dependencia de artefactos estáticos y lineales (sílabos) que ocultan la topología real de las competencias y sus dependencias causales. Esta fragmentación genera una desalineación pedagógica crítica y "puntos ciegos" en el andamiaje del aprendizaje, dificultando la monitorización del progreso estudiantil y el cumplimiento de estándares nacionales e internacionales de calidad. Este artículo presenta Skill Tree Studio (STS), un sistema innovador que implementa un pipeline de Sistemas Multi-Agente (MAS) basado en modelos de lenguaje de gran escala (LLMs) para la deconstrucción ontológica automática de currículos. A diferencia de los enfoques heurísticos tradicionales, STS orquesta una secuencia determinista de agentes especializados: un analista de contexto disciplinar, un extractor de metadatos taxonómicos (Bloom) y un arquitecto de grafos topológicos. El framework utiliza inferencia semántica de "arranque en frío" para transformar texto no estructurado en un grafo dirigido acíclico (DAG) de habilidades, garantizando integridad matemática y coherencia pedagógica mediante esquemas JSON forzados. Demostramos cómo esta estructura formal habilita la analítica curricular proactiva, funcionando como un co-piloto para el educador ("Teacher in the loop"). El panel de analíticas traduce métricas de la teoría de grafos (como la ruta crítica de aprendizaje, la centralidad de intermediación y la taxonomía de Bloom) en insights accionables, permitiendo la detección temprana de habilidades clave (Keystone Skills) y la prevención de cuellos de botella conceptuales donde los estudiantes pueden llegar a fallar. Asimismo, la topología del grafo sirve como contexto fundacional (GraphRAG) para la generación automatizada de evaluaciones psicométricamente alineadas y recursos de aprendizaje adaptativos mejorando la experiencia del alumno. El sistema no está ajustado a un dominio particular. El mismo formalismo DAG + pipeline multi-agente funciona sobre estructuras epistemológicas muy distintas. Los hallazgos sugieren que la ingeniería curricular aumentada de STS reduce significativamente la carga cognitiva del diseño instruccional, permitiendo una transición de la gestión documental a la optimización sistémica del aprendizaje. El sistema se valida como una herramienta de alto impacto para la acreditación institucional, alineándose con marcos regulatorios exigentes (como el Decreto 1330/2019 y el Modelo CNA 2025 en Colombia)[1-2], al garantizar la pertinencia, la transparencia estructural y la evidencia objetiva del diseño pedagógico basado en resultados de aprendizaje.
dc.description.abstractenglish	Contemporary curriculum architecture is increasingly affected by what may be termed structural opacity: a reliance on static, linear artifacts (syllabi) that obscure the true topology of competencies and their causal dependencies. This fragmentation generates critical pedagogical misalignment and conceptual “blind spots” within the learning scaffold, hindering effective monitoring of student progress and compliance with rigorous national and international quality standards. This article introduces Skill Tree Studio (STS), an innovative system that implements a Multi-Agent Systems (MAS) pipeline grounded in Large Language Models (LLMs) for the automatic ontological deconstruction of curricula. Unlike traditional heuristic approaches, STS orchestrates a deterministic sequence of specialized agents: a disciplinary context analyst, a taxonomic metadata extractor (Bloom’s Taxonomy), and a topological graph architect. The framework employs cold-start semantic inference to transform unstructured textual syllabi into a Directed Acyclic Graph (DAG) of skills, ensuring mathematical integrity and pedagogical coherence through enforced JSON schemas. We demonstrate how this formal structure enables proactive curriculum analytics, functioning as a “teacher-in-the-loop” co-pilot. The analytics dashboard translates graph-theoretical metrics—such as learning critical path, betweenness centrality, and Bloom-level distribution—into actionable insights. This allows early identification of Keystone Skills and the prevention of conceptual bottlenecks where learners are at risk of failure. Furthermore, the graph topology serves as foundational context (GraphRAG) for the automated generation of psychometrically aligned assessments and adaptive learning resources, thereby enhancing the overall learner experience. The system is domain-agnostic: the same DAG formalism and multi-agent pipeline operate effectively across diverse epistemological structures. Findings suggest that STS-augmented curriculum engineering significantly reduces the cognitive load associated with instructional design, enabling a shift from document management toward systemic learning optimization. The system is validated as a high-impact tool for institutional accreditation, aligning with stringent regulatory frameworks (such as Decree 1330/2019 and the CNA 2025 Model in Colombia), by ensuring relevance, structural transparency, and objective evidence of outcome-based pedagogical design.
dc.format.extent	59 páginas
dc.format.mimetype	application/pdf
dc.identifier.uri	https://hdl.handle.net/20.500.12010/39310
dc.language.iso	es
dc.relation.references	Ministerio de Educación Nacional de Colombia, "Decreto 1330 de 2019: Por el cual se sustituye el Capítulo 2 y se suprime el Capítulo 7 del Título 3 de la Parte 5 del Libro 2 del Decreto 1075 de 2015 —Único Reglamentario del Sector Educación," Bogotá, D.C., Colombia, jul. 25, 2019.
dc.relation.references	Consejo Nacional de Educación Superior (CESU), "Acuerdo 01 de 2025: Lineamientos y aspectos por evaluar para la acreditación en alta calidad de los programas académicos, las unidades académicas y las instituciones de educación superior," Bogotá, D.C., Colombia, dic. 16, 2025.
dc.relation.references	P. R. Aldrich, “The curriculum prerequisite network: Modeling the curriculum as a complex system,” Biochem. Mol. Biol. Educ., vol. 43, no. 3, pp. 168–180, May 2015
dc.relation.references	P. Stavrinides and K. M. Zuev, “Course-prerequisite networks for analyzing and understanding academic curricula,” Appl. Netw. Sci., vol. 8, no. 1, p. 19, Apr. 2023.
dc.relation.references	J. M. Lightfoot, “A Graph-Theoretic Approach to Improved Curriculum Structure and Assessment Placement,” 2023
dc.relation.references	A. Slim, C. Abdallah, E. Allen, and A. Slim, “Integrated Curriculum Analytics: Bridging Structure, Pass Rates, and Student Outcomes,” arXiv preprint, 2025.
dc.relation.references	H. Bijl, “Structuring Competency-Based Courses Through Skill Trees,” Koli Calling ’25, 2025.
dc.relation.references	M. R. Frank, L. Sun, B. Al Shebli, C. Hidalgo, and I. Rahwan, “Unpacking the polarization of workplace skills”, Science Advances, vol. 4, no. 7, p., 2018.
dc.relation.references	D. Varagnolo, S. Knorn, K. Staffas, E. Fjällström, and T. Wrigstad, “Graph-theoretic approaches and tools for quantitatively assessing curricula coherence,” Eur. J. Eng. Educ., vol. 46, no. 3, pp. 344–363, May 2021.
dc.relation.references	B. Yang et al., “Analysis of Student Progression Through Curricular Networks: A Case Study in an Illinois Public Insti,” 2025.
dc.relation.references	L. W. Anderson y D. R. Krathwohl, Eds., A Taxonomy for Learning, Teaching, and Assessing: A Revision of Bloom's Taxonomy of Educational Objectives. New York, NY, USA: Longman, 2001.
dc.relation.references	C. Simon de Blas, D. Gomez Gonzalez, and R. Criado Herrero, “Network analysis: An indispensable tool for curricula design. A real case-study of the degree on mathematics at the URJC in Spain,” PLoS One, vol. 16, no. 3, p. 1, Mar. 2021.
dc.relation.references	B. Abu-Salih and S. Alotaibi, “A systematic literature review of knowledge graph construction and application in education,” Heliyon, vol. 10, no. 3, 1, Feb. 2024.
dc.relation.references	X. Yu, H. Chen, M. Stahr, and R. Yan, “Design and Implementation of Curriculum System Based on Knowledge Graph,” 2020.
dc.relation.references	Curriculum, Skill Trees, Knowledge Graphs, Multi-Agent Systems, Large Language Models (LLMs), GraphRAG, Curriculum Analytics, Curriculum Engineering.
dc.relation.references	R. Manrique, B. Pereira, and O. Mariño, “Exploring knowledge graphs for the identification of concept prerequisites,” Smart Learn. Environ., vol. 6, no. 1, p. 21, 2019.
dc.relation.references	X. Yu et al., “Design and Implementation of Curriculum System Based on Knowledge Graph,” 2020.
dc.relation.references	P. Chen, Y. Lu, V. W. Zheng, X. Chen, and B. Yang, “KnowEdu: A System to Construct Knowledge Graph for Education,” IEEE Access, vol. 6, pp. 31553–31563, 2018.
dc.relation.references	M. C. Aytekin and Y. Saygın, “ACE: AI-assisted Construction of Educational Knowledge Graphs with Prerequisite Relations,” J. Educ. Data Min., vol. 16, no. 2, pp. 85–114, 2024.
dc.relation.references	T. Yang, B. Ren, C. Gu, T. He, B. Ma, and S. Konomi, “Examining GPT’s Capability to Generate and Map Course Concepts and Their Relationships,” 2025.
dc.relation.references	T. Yang, B. Ren, C. Gu, T. He, B. Ma, and S. Konomi, “Leveraging LLMs for Automated Extraction and Structuring of Educational Concepts and Relationships,” Mach. Learn. Knowl. Extr., vol. 7, no. 3, p. 103, Sep. 2025.
dc.relation.references	Q. Wu et al., "AutoGen: Enabling Next-Gen LLM Applications via Multi-Agent Conversation," arXiv preprint arXiv:2308.08155, 2023 (p. 1).
dc.relation.references	X. Zhang et al., “EduPlanner: LLM-Based Multi-Agent Systems for Customized and Intelligent Instructional Design,” 2024.
dc.relation.references	P. Chen, Y. Lu, V. W. Zheng, X. Chen y B. Yang, «KnowEdu: A System to Construct Knowledge Graph for Education», IEEE Access, vol. 6, pp. 31553–31563, may 2018. (pp. 1, 11).
dc.relation.references	G. Konidaris, S. Kuindersma, A. Barto y R. Grupen, «Constructing skill trees for reinforcement learning agents from demonstration trajectories», en Proc. 24th Int. Conf. Neural Inf. Process. Syst., Vancouver, BC, Canada, 2010, pp. 1162-1170.
dc.relation.references	D. Reales, R. Manrique, and C. Grévisse, “Core concept identification in educational resources via knowledge graphs and large language models,” SN Comput. Sci., vol. 5, no. 8, p. 1029, 2024.
dc.relation.references	H. Tebourbi et al., “BPMN-Based Design of Multi-Agent Systems: Personalized Language Learning Workflow Automation with RAG-Enhanced Knowledge Access,” Information, vol. 16, no. 9, p. 809, Sep. 2025.
dc.relation.references	J. Grosch and H. Emmelmann, “A tool box for compiler construction,” in Compiler Compilers: Third International Workshop, CC’90 Schwerin, FRG, October 22–24, 1990 Proceedings 3, Springer, 1991, pp. 106–116.
dc.relation.references	L. Huang et al., “A Survey on Hallucination in Large Language Models: Principles, Taxonomy, Challenges, and Open Questions,” arXiv e-prints, p. arXiv:2311.05232, 2023.
dc.relation.references	Z. Chu et al., “LLM Agents for Education: Advances and Applications,” 2025.
dc.relation.references	J. Luo et al., “Large Language Model Agent: A Survey on Methodology, Applications and Challenges,” 2025.
dc.relation.references	A. T. Aliaga Torró, “Towards a modular and adaptive AI architecture for personalized intelligent tutoring systems,” Master’s Thesis, UNIVERSITAT POLITÈCNICA DE VALÈNCIA, 2025
dc.relation.references	A. S. M. M. Hasan, K. B. Shahnoor, and S. S. Tasneem, “Automatic Question & Answer Generation Using Generative Large Language Model (LLM),” 2025.
dc.relation.references	L. Huang, W. Yu, W. Ma, W. Zhong, Z. Feng, H. Wang, Q. Chen, W. Peng, X. Feng, B. Qin, and T. Liu, “A Survey on Hallucination in Large Language Models: Principles, Taxonomy, Challenges, and Open Questions,” arXiv e-prints, p. arXiv:2311.05232, 2023.
dc.relation.references	D. Varagnolo et al., “Graph-theoretic approaches and tools for quantitatively assessing curricula coherence,” Eur. J. Eng. Educ., vol. 46, no. 3, pp. 344–363, May 2021.
dc.relation.references	A. Slim et al., “Integrated Curriculum Analytics: Bridging Structure, Pass Rates, and Student Outcomes,” arXiv preprint, 2025.
dc.relation.references	P. Stavrinides and K. M. Zuev, “Course-prerequisite networks for analyzing and understanding academic curricula,” Appl. Netw. Sci., vol. 8, no. 1, p. 19, Apr. 2023.
dc.relation.references	Z. Liu et al., “AcademicRAG Knowledge Graph Enhanced Retrieval-Augmented Generation for Academic Resource Discovery,” 2025.
dc.relation.references	Y. Wang et al., “Multi-Examiner: A Knowledge Graph-Driven System for Generating Comprehensive IT Questions with Higher-Order Thinking,” Appl. Sci., vol. 15, no. 10, p. 5719, May 2025.
dc.relation.references	D. Olaniyan, J. Olaniyan, I. C. Obagbuwa, and A. K. Tsetse, “eXplainable AI Framework for Automated Lesson Plan Generation and Alignment with Bloom’s Taxonomy,” Computers, vol. 14, no. 11, p. 494, Nov. 2025
dc.relation.references	B. Yang et al., “Analysis of Student Progression Through Curricular Networks A Case Study in an Illinois Public Insti,” 2025.
dc.subject	Currículo
dc.subject	Gráficos de Conocimiento
dc.subject	Sistemas Multi-Agente
dc.subject	LLMs
dc.subject	GraphRAG
dc.subject	Analítica Curricular
dc.subject	Ingeniería Curricular
dc.subject.keyword	Curriculum
dc.subject.keyword	Skill Trees
dc.subject.keyword	Knowledge Graphs
dc.subject.keyword	Multi-Agent Systems
dc.subject.keyword	Large Language Models (LLMs)
dc.subject.keyword	GraphRAG
dc.subject.keyword	Curriculum Analytics
dc.subject.keyword	Curriculum Engineering
dc.subject.lemb	Diseño curricular - Innovaciones tecnológicas
dc.subject.lemb	Inteligencia artificial - Aplicaciones en educación
dc.subject.lemb	Sistemas multiagente (Inteligencia artificial).
dc.title	From linear syllabi to dynamic knowledge graphs: an llm-powered multi-agent system for co-creative curriculum design and proactive analysis
dc.type.coar	http://purl.org/coar/resource_type/c_bdcc

Archivos

Bloque original

Mostrando 1 - 1 de 1

Nombre:: From Linear Syllabi to Dynamic Knowledge Graphs (Tesis).pdf
Tamaño:: 40.16 MB
Formato:: Adobe Portable Document Format

Descargar

Bloque de licencias

Mostrando 1 - 2 de 2

Nombre:: license.txt
Tamaño:: 3.28 KB
Formato:: Item-specific license agreed upon to submission
Descripción:

Descargar

Nombre:: FOR-EFE-GDB-008_AUTORIZACION_DE_PUBLICACION_DE_TESIS_O_TRABAJO_DE_GRADO_DE_FORMA_CONFIDENCIAL (6)_Nelson.pdf
Tamaño:: 232.61 KB
Formato:: Adobe Portable Document Format
Descripción:: Carta de autorización

Descargar

Colecciones

Maestría en Modelado y Simulación MM&S