Elamur Research⌕
☰
home / papers / 2026 / ARAMIS: a web tool integrated with an open-source LLM to support grading undergraduate theses

ARAMIS: a web tool integrated with an open-source LLM to support grading undergraduate theses

Original (Portuguese): ARAMIS: Uma ferramenta web integrada com LLM open-source para apoio à correção de TCCs de estudantes de graduação

Paper · 2026 · Gustavo Campelo de Sousa, Pablo Kauan Martins Timbó, Luiz Zairo Bastos Viana et al.

Summary

ARAMIS (Academic Review Agents for Methodological Improvements) is an open-source web tool that helps undergraduates in Brazil revise their final-year theses (TCC) written in Portuguese. The authors argue that most LLM-based reviewing systems target English and mainly fix surface errors. ARAMIS also looks at argument coherence and methodological soundness. Its multi-agent architecture is orchestrated with the Agno framework, and every agent calls the open-source gpt-oss-20b model through the Hugging Face API. There are three agents: grammar correction, logical chaining, and methodological rigor. Users fill in configuration fields, paste a text excerpt, and receive structured feedback for whichever agents they select. Past reviews are saved in a personal history. The evaluation covered usability only, using the System Usability Scale with ten participants. The tool scored 90.5/100, which falls in the top band of the Bangor et al. scale. The quality of the generated reviews was not measured. The work matters because it brings LLM-assisted academic feedback, beyond spell-checking, to Portuguese-speaking students and could ease the supervision workload.

Key points

  • Architecture: a web app with three Agno-orchestrated agents, all backed by gpt-oss-20b via the Hugging Face API.
    • The grammar agent flags spelling and accent errors.
    • The logical chaining agent checks coherence between sentences.
    • The methodological rigor agent checks that the research problem, objectives, method and results analysis are consistent.
  • Prompting: prompts use few-shot examples and contain placeholders filled from the user's input. The variables are secao_desejada, titulo_tcc, area_conhecimento_tcc and nivel_rigor_modelo (the strictness level).
  • Usability result: SUS score of 90.5/100, rated "Excellent" on the Bangor et al. (2009) scale. Participants praised its simplicity and the accuracy of the grammar agent in particular.
  • Limitations:
    • Only ten users from one institution took part.
    • The quality of the reviews themselves was not evaluated.
    • Only a narrow range of models was tried.
    • The free API is rate-limited, and the hosting infrastructure cannot run the model locally.
  • Future work: comparing candidate models, fine-tuning an open-source LLM, assessing review quality, testing with a larger user base, and self-hosting to remove the dependence on external APIs.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Fields, topics and concepts

📚 Shelves 2The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 works

🧭 Topics 1Research subjects and languages the work is about.Language models & pretraining · 3,320 works

Cite

Cite as Gustavo Campelo de Sousa, Pablo Kauan Martins Timbó, Luiz Zairo Bastos Viana, Antônio Emerson Barros Tomaz, José Wellington Franco da Silva and Carlos de Oliveira Caminha. 2026. ARAMIS: Uma ferramenta web integrada com LLM open-source para apoio à correção de TCCs de estudantes de graduação. Proceedings of the 17th International Conference on Computational Processing of Portuguese (PROPOR 2026) - Vol. 2.

The paper itself, not this page. BibTeX from the library's record of the work.

@inproceedings{sousa2026aramis,
  title = {{ARAMIS: Uma ferramenta web integrada com LLM open-source para apoio à correção de TCCs de estudantes de graduação}},
  author = {Gustavo Campelo {de Sousa}, Pablo Kauan Martins Timbó, Luiz Zairo Bastos Viana, Antônio Emerson Barros Tomaz, José Wellington Franco {da Silva} and Carlos {de Oliveira Caminha}},
  booktitle = {Proceedings of the 17th International Conference on Computational Processing of Portuguese (PROPOR 2026) - Vol. 2},
  year = {2026},
  url = {https://aclanthology.org/2026.propor-2.1/}
}

References

16 references, 0 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Maximilian Idahl and Zahra Ahmadi. 2025. OpenReviewer: A specialized large language model for generating critical scientific paper reviews. In Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Syst… find
  2. Associação Brasileira de Normas Técnicas. 2024. NBR 14724: Informação e documentação – Trabalhos acadêmicos – Apresentação. Rio de Janeiro: ABNT. 4. ed. find
  3. Aaron Bangor, Philip Kortum, and James Miller. 2009. Determining what individual sus scores mean: Adding an adjective rating scale. Journal of usability studies, 4(3):114–123. find
  4. Ankur Joshi, Saket Kale, Satish Chandel, and D Kumar Pal. 2015. Likert scale: Explored and explained. British journal of applied science & technology, 7(4):396. find
  5. Zairo Bastos, Carlos Caminha, and Wellington Franco. 2025a. Llm4time: Uma ferramenta interativa para previsão de séries temporais com modelos largos de linguagem. In Anais Estendidos do XL Simpósio Brasileiro de Bancos de Dados, pages 100–105, Porto Alegre, RS, Brasil. SBC. find
  6. Weixin Liang, Yuhui Zhang, Hancheng Cao, Binglu Wang, Daisy Yi Ding, Xinyu Yang, Kailas Vodrahalli, Siyu He, Daniel Scott Smith, Yian Yin, and 1 others. 2024. Can large language models provide useful feedback on research papers? a large-scale empirical analysis. NEJM AI, 1(8):AIoa2400196. find
  7. Zairo Bastos, João David Freitas, José Wellington Franco da Silva, and Carlos Caminha. 2025b. Prompt-driven time series forecasting with large language models. In ICEIS (1), pages 309–316. find
  8. Andrea A. Lunsford and Karen J. Lunsford. 2008. "mistakes are a fact of life": A national comparative study. College Composition and Communication, 59(4):781– 806. find
  9. Zairo Bastos, Raylander Marques, Gabriel Rudan, Marlon Duarte, and Wellington Franco. 2025c. Obi-uan: Um agente para auxiliar nos estudos da olimpíada brasileira de informática. In Anais Estendidos do XL Simpósio Brasileiro de Bancos de Dados, pages 94–99, Porto Alegre, RS, Brasil. SBC. find
  10. Wesley Barbosa Silva, Maria Fernanda Aquino Freitas Scarcela, Luiz Zairo Bastos Viana, Carlos Caminha, João Paulo do Vale Madeiro, and José Wellington Franco da Silva. 2026. LLM4series: Structured prompting for time series forecasting with LLMs. In 1st ICLR Workshop on Time Series in the Age of Larg… find
  11. John Brooke. 1995. Sus: A quick and dirty usability scale. Usability Eval. Ind., 189. find
  12. Rosadélia Malheiros Carboni and Valnice de Oliveira Nogueira. 2008. Facilidades e dificuldades na elaboração de trabalhos de conclusão de curso. ConScien-tiae Saúde, 3:65–72. find
  13. Nalla Srivarsha, Gummadi Nithin, Shanmugasundaram Hariharan, Bibhuti Bhusan Dash, Subrata Chowdhury, and Sudhansu Shekhar Patra. 2025. Auto text correction using nlp techniques. In 2025 6th International Conference on Mobile Computing and Sustainable Informatics (ICMCSI), pages 590–594. IEEE. find
  14. Eric Chamoun, Michael Schlichtkrull, and Andreas Vlachos. 2024. Automated focused feedback generation for scientific writing assistance. In Findings of the Association for Computational Linguistics: ACL 2024, pages 9742–9763, Bangkok, Thailand. Association for Computational Linguistics. find
  15. Wenhai Wang, Zhe Chen, Xiaokang Chen, Jiannan Wu, Xizhou Zhu, Gang Zeng, Ping Luo, Tong Lu, Jie Zhou, Yu Qiao, and 1 others. 2024. Visionllm: Large language model is also an open-ended decoder for vision-centric tasks. Advances in Neural Information Processing Systems, 36. find
  16. Mike D’Arcy, Tom Hope, Larry Birnbaum, and Doug Downey. 2024. Marg: Multi-agent review generation for scientific papers. Preprint, arXiv:2401.04259. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/sousa-2026-aramis-uma-ferramenta-web-integrada/