EduQuery: Consulta Confiável a Indicadores Educacionais via Sistema Multi-Agente com Guardrails contra Alucinações
Resumo
O acesso a indicadores educacionais comparativos é dificultado pela heterogeneidade das fontes oficiais. Modelos de Linguagem com recuperação permitem consultas em linguagem natural, mas alucinações inviabilizam decisões fundamentadas. Apresenta-se o EduQuery, assistente multi-agente que organiza dados de fontes oficiais, com guardrails determinísticos como contribuição central. Avaliaram-se 84 consultas. Nos indicadores modelados, o sistema interceptou 55,6% das alucinações que passariam sem verificação e elevou a acurácia de 10% para 60% (média de 63,3%, desvio padrão de 11,5% em três repetições); nas 30 adversariais, o comportamento correto foi de 83,3%. O EduQuery independe do provedor do Modelo de Linguagem.
Palavras-chave:
Educação Comparada, Sistemas Multi-Agente, Guardrails Determinísticos, Data Lakehouse Educacional, Reprodutibilidade
Referências
Angrist, N., Djankov, S., Goldberg, P. K., and Patrinos, H. A. (2021). Measuring human capital using global learning data. Nature, 592:403–408. DOI: 10.1038/s41586-021-03323-7.
Armbrust, M., Ghodsi, A., Xin, R., and Zaharia, M. (2021). Lakehouse: A new generation of open platforms that unify data warehousing and advanced analytics. In Conference on Innovative Data Systems Research (CIDR).
Asai, A., Wu, Z., Wang, Y., Sil, A., and Hajishirzi, H. (2024). Self-RAG: Learning to retrieve, generate, and critique through self-reflection. In International Conference on Learning Representations (ICLR).
Basili, V. R., Caldiera, G., and Rombach, H. D. (1994). The goal question metric approach. In Encyclopedia of Software Engineering, pages 528–532. John Wiley & Sons.
Carnoy, M., Khavenson, T., Fonseca, I., Costa, L., and Marotta, L. (2015). A educação brasileira está melhorando? evidências do PISA e do SAEB. Cadernos de Pesquisa, 45(157):450–485. DOI: 10.1590/198053143331.
Edge, D., Trinh, H., Cheng, N., Bradley, J., Chao, A., Mody, A., Truitt, S., and Larson, J. (2024). From local to global: A graph RAG approach to query-focused summarization. Disponível em: [link]. Acesso em: 14 jul. 2026.
Fernandes, R. (2007). Índice de desenvolvimento da educação básica (IDEB). Série Documental 26, Instituto Nacional de Estudos e Pesquisas Educacionais Anísio Teixeira (INEP).
Guardrails AI (2024). Guardrails AI: Adding guardrails to large language models. Disponível em: [link]. Acesso em: 14 jul. 2026.
Hanushek, E. A. and Woessmann, L. (2011). The economics of international differences in educational achievement. In Hanushek, E. A., Machin, S., and Woessmann, L., editors, Handbook of the Economics of Education, volume 3, pages 89–200. Elsevier.
Ji, Z., Lee, N., Frieske, R., Yu, T., Su, D., Xu, Y., Ishii, E., Bang, Y. J., Madotto, A., and Fung, P. (2023). Survey of hallucination in natural language generation. ACM Computing Surveys, 55(12). DOI: 10.1145/3571730.
LangChain (2024). LangGraph: Building stateful, multi-actor applications with LLMs. Disponível em: [link]. Acesso em: 14 jul. 2026.
Lewis, P., Perez, E., Piktus, A., Petroni, F., Karpukhin, V., Goyal, N., Küttler, H., Lewis, M., Yih, W., Rocktäschel, T., Riedel, S., and Kiela, D. (2020). Retrieval-augmented generation for knowledge-intensive NLP tasks. In Advances in Neural Information Processing Systems (NeurIPS), volume 33.
Mullis, I. V. S., von Davier, M., Foy, P., Fishbein, B., Reynolds, K. A., and Wry, E. (2023). PIRLS 2021 international results in reading. Technical report, TIMSS & PIRLS International Study Center, Boston College.
OECD (2024). Education at a glance 2024: OECD indicators. OECD Publishing, Paris. DOI: 10.1787/c00cad36-en.
Raasveldt, M. and Mühleisen, H. (2019). DuckDB: an embeddable analytical database. In Proceedings of the 2019 International Conference on Management of Data (SIGMOD), pages 1981–1984. DOI: 10.1145/3299869.3320212.
Rebedea, T., Dinu, R., Sreedhar, M., Parisien, C., and Cohen, J. (2023). NeMo Guardrails: A toolkit for controllable and safe LLM applications with programmable rails. In Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing (EMNLP): System Demonstrations.
Schleicher, A. (2018). World Class: How to Build a 21st-Century School System. OECD Publishing, Paris. DOI: 10.1787/9789264300002-en.
UNESCO (2020). Global education monitoring report 2020: Inclusion and education — all means all. UNESCO Publishing.
Wohlin, C., Runeson, P., Höst, M., Ohlsson, M. C., Regnell, B., and Wesslén, A. (2012). Experimentation in Software Engineering. Springer, 2nd edition.
Wu, Q., Bansal, G., Zhang, J., Wu, Y., Li, B., Zhu, E., Jiang, L., Zhang, X., Zhang, S., Liu, J., Awadallah, A. H., White, R. W., Burger, D., and Wang, C. (2024). AutoGen: Enabling next-gen LLM applications via multi-agent conversation. In Proceedings of the First Conference on Language Modeling (COLM).
Yan, S.-Q., Gu, J.-C., Zhu, Y., and Ling, Z.-H. (2024). Corrective retrieval augmented generation. Disponível em: [link]. Acesso em: 14 jul. 2026.
Zheng, L., Chiang, W.-L., Sheng, Y., Zhuang, S., Wu, Z., Zhuang, Y., Lin, Z., Li, Z., Li, D., Xing, E. P., Zhang, H., Gonzalez, J. E., and Stoica, I. (2023). Judging LLM-as-a-judge with MT-Bench and Chatbot Arena. In Advances in Neural Information Processing Systems (NeurIPS), volume 36.
Armbrust, M., Ghodsi, A., Xin, R., and Zaharia, M. (2021). Lakehouse: A new generation of open platforms that unify data warehousing and advanced analytics. In Conference on Innovative Data Systems Research (CIDR).
Asai, A., Wu, Z., Wang, Y., Sil, A., and Hajishirzi, H. (2024). Self-RAG: Learning to retrieve, generate, and critique through self-reflection. In International Conference on Learning Representations (ICLR).
Basili, V. R., Caldiera, G., and Rombach, H. D. (1994). The goal question metric approach. In Encyclopedia of Software Engineering, pages 528–532. John Wiley & Sons.
Carnoy, M., Khavenson, T., Fonseca, I., Costa, L., and Marotta, L. (2015). A educação brasileira está melhorando? evidências do PISA e do SAEB. Cadernos de Pesquisa, 45(157):450–485. DOI: 10.1590/198053143331.
Edge, D., Trinh, H., Cheng, N., Bradley, J., Chao, A., Mody, A., Truitt, S., and Larson, J. (2024). From local to global: A graph RAG approach to query-focused summarization. Disponível em: [link]. Acesso em: 14 jul. 2026.
Fernandes, R. (2007). Índice de desenvolvimento da educação básica (IDEB). Série Documental 26, Instituto Nacional de Estudos e Pesquisas Educacionais Anísio Teixeira (INEP).
Guardrails AI (2024). Guardrails AI: Adding guardrails to large language models. Disponível em: [link]. Acesso em: 14 jul. 2026.
Hanushek, E. A. and Woessmann, L. (2011). The economics of international differences in educational achievement. In Hanushek, E. A., Machin, S., and Woessmann, L., editors, Handbook of the Economics of Education, volume 3, pages 89–200. Elsevier.
Ji, Z., Lee, N., Frieske, R., Yu, T., Su, D., Xu, Y., Ishii, E., Bang, Y. J., Madotto, A., and Fung, P. (2023). Survey of hallucination in natural language generation. ACM Computing Surveys, 55(12). DOI: 10.1145/3571730.
LangChain (2024). LangGraph: Building stateful, multi-actor applications with LLMs. Disponível em: [link]. Acesso em: 14 jul. 2026.
Lewis, P., Perez, E., Piktus, A., Petroni, F., Karpukhin, V., Goyal, N., Küttler, H., Lewis, M., Yih, W., Rocktäschel, T., Riedel, S., and Kiela, D. (2020). Retrieval-augmented generation for knowledge-intensive NLP tasks. In Advances in Neural Information Processing Systems (NeurIPS), volume 33.
Mullis, I. V. S., von Davier, M., Foy, P., Fishbein, B., Reynolds, K. A., and Wry, E. (2023). PIRLS 2021 international results in reading. Technical report, TIMSS & PIRLS International Study Center, Boston College.
OECD (2024). Education at a glance 2024: OECD indicators. OECD Publishing, Paris. DOI: 10.1787/c00cad36-en.
Raasveldt, M. and Mühleisen, H. (2019). DuckDB: an embeddable analytical database. In Proceedings of the 2019 International Conference on Management of Data (SIGMOD), pages 1981–1984. DOI: 10.1145/3299869.3320212.
Rebedea, T., Dinu, R., Sreedhar, M., Parisien, C., and Cohen, J. (2023). NeMo Guardrails: A toolkit for controllable and safe LLM applications with programmable rails. In Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing (EMNLP): System Demonstrations.
Schleicher, A. (2018). World Class: How to Build a 21st-Century School System. OECD Publishing, Paris. DOI: 10.1787/9789264300002-en.
UNESCO (2020). Global education monitoring report 2020: Inclusion and education — all means all. UNESCO Publishing.
Wohlin, C., Runeson, P., Höst, M., Ohlsson, M. C., Regnell, B., and Wesslén, A. (2012). Experimentation in Software Engineering. Springer, 2nd edition.
Wu, Q., Bansal, G., Zhang, J., Wu, Y., Li, B., Zhu, E., Jiang, L., Zhang, X., Zhang, S., Liu, J., Awadallah, A. H., White, R. W., Burger, D., and Wang, C. (2024). AutoGen: Enabling next-gen LLM applications via multi-agent conversation. In Proceedings of the First Conference on Language Modeling (COLM).
Yan, S.-Q., Gu, J.-C., Zhu, Y., and Ling, Z.-H. (2024). Corrective retrieval augmented generation. Disponível em: [link]. Acesso em: 14 jul. 2026.
Zheng, L., Chiang, W.-L., Sheng, Y., Zhuang, S., Wu, Z., Zhuang, Y., Lin, Z., Li, Z., Li, D., Xing, E. P., Zhang, H., Gonzalez, J. E., and Stoica, I. (2023). Judging LLM-as-a-judge with MT-Bench and Chatbot Arena. In Advances in Neural Information Processing Systems (NeurIPS), volume 36.
Publicado
05/10/2026
Como Citar
RIBEIRO, Tharsys da Silva; VASCONCELOS, Igor Oliveira.
EduQuery: Consulta Confiável a Indicadores Educacionais via Sistema Multi-Agente com Guardrails contra Alucinações. In: SIMPÓSIO BRASILEIRO DE INFORMÁTICA NA EDUCAÇÃO (SBIE), 37. , 2026, Goiânia/GO.
Anais [...].
Porto Alegre: Sociedade Brasileira de Computação,
2026
.
p. 2345-2359.
DOI: https://doi.org/10.5753/sbie.2026.28418.
