DOI: 10.17587/prin.17.334-344

Platform for Automated Creation of Individual Educational Trajectories based on LLM and RAG

V. S. Andreev, Master's Student, viktormoonwell123@yandex.ru, A. G. Kravets, D. Sc. (Eng.), Professor, AllaGKravets@yandex.ru, N. S. Kachurin, Master's Student, ka4urin.nik@yandex.ru, Volgograd State Technical University, Volgograd, 400005, Russian Federation

Corresponding author: Viktor S. Andreev, Master's Student, Volgograd State Technical University, Volgograd, 400005, Russian Federation, E-mail: viktormoonwell123@yandex.ru

This paper presents a model and a prototype of a software platform for the automated formation of an individual educational trajectory (IET) based on Large Language Models (LLM) and Retrieval-Augmented Generation (RAG). An analytical review of international and Russian digital educational platforms is conducted, demonstrating that existing solutions are primarily focused on navigation and recommendation of individual courses or programs and often rely on multi-stage testing procedures and organizational mediation. These characteristics significantly limit the flexibility and operational efficiency of IET construction, especially in scenarios where rapid personalization is required. The proposed platform adopts a fundamentally different approach by using a free-text description of a talent profile as the primary input. The input data are automatically structured into a JSON format and used to generate a six-month individual educational trajectory. To improve the grounding, consistency, and relevance of recommendations, a dedicated corpus of educational courses was constructed using materials from the Stepik platform and integrated into the generation pipeline via a RAG module. An experimental evaluation was performed on a dataset of 1,000 talent profiles generated under controlled conditions. The quality of the generated IETs was assessed using the LLM-as-a-judge methodology, with an independent evaluation model. Several generation modes were compared, including LLM-only generation and multiple retrieval-based strategies. The results show that the integration of RAG increases IET generation accuracy from 80 % (LLM-only) to 82.28 %. At the same time, simple retrieval strategies, such as Random, TF-IDF, BM25, and Dense Retrieval, were found to reduce overall quality compared to the LLM-only baseline.

Keywords: artificial intelligence, LLM, RAG, individual educational trajectory, personalization, digital platform, recommendations, system architecture, JSON data, API

pp. 334—344

For citation:
Andreev V. S., Kravets A. G., Kachurin N. S. Platform for Automated Creation of Individual Educational Trajectories based on LLM and RAG, Programmnaya Ingeneria, 2026, vol. 17, no. 6, pp. 334—344. DOI: 10.17587/prin.17.334-344 (in Russian).

References:

1. Meyers M. C., van Woerkom M. The Influence of Underlying Philosophies on Talent Management: Theory, Implications for Practice, and Research Agenda, Journal of World Business, 2014, vol. 49, no. 2, pp. 192—203. DOI: 10.1016/j.jwb.2013.11.003.
2. Huang Y., Huang J. A survey on Retrieval-Augmented Text Generation for Large Language Models, arXiv, 2024. DOI: 10.48550/arXiv.2404.10981.
3. Gupta S., Ranjan R., Singh S. N. A Comprehensive Survey of Retrieval-Augmented Generation (RAG): Evolution, Current Landscape and Future Directions, arXiv, 2024. DOI: 10.48550/arXiv.2410.12837.
4. Lewis P., Perez E., Piktus A. et al. Retrieval-Augmented Generation for Knowledge-Intensive NLP tasks, arXiv, 2021. DOI: 10.48550/arXiv.2005.11401.
5. Izacard G., Lewis P., Lomeli M. et al. Atlas: Few-shot Learning with Retrieval Augmented Language Models, arXiv, 2022. DOI: 10.48550/arXiv.2208.03299.
6. Borgeaud S., Mensch A., Hoffmann J. et al. Improving Language Models by Retrieving from Trillions of Tokens, arXiv, 2022. DOI: 10.48550/arXiv.2112.04426.
7. Wu L., Zheng Z., Qiu Z. et al. A Survey on Large Language Models for Recommendation, arXiv, 2024. DOI: 10.48550/arXiv.2305.19860.
8. Rahayu N. W., Ferdiana R., Kusumawardani S. A Systematic Review of Learning Path Recommender Systems, Education and Information Technologies, 2023, vol. 28, no. 6, pp. 7437—7460.DOI: 10.1007/s10639-022-11460-3.
9. Valcarce D., Bellogin A., Parapar J., Castells P. Assessing Ranking Metrics in Top-N Recommendation, Information Retrieval Journal, 2020, vol. 23, pp. 411—448. DOI: 10.1007/s10791-020-09377-x.
10. Niknam M., Thulasiraman P. LPR: A Bio-Inspired Intelligent Learning Path Recommendation System Based on Meaningful Learning Theory, Education and Information Technologies, 2020, vol. 25, no. 5, pp. 3797—3819. DOI: 10.1007/s10639-020-10133-3.
11. Tzeng J. W., Huang N. F., Chuang A. C. et al. Massive Open Online Course Recommendation System based on a Reinforcement Learning Algorithm, Neural Computing and Applications, 2025, vol. 37, pp. 11607—11618. DOI: 10.1007/s00521-023-08686-8.
12. Elsom-Cook M. Student Modelling in Intelligent Tutoring Systems, Artificial Intelligence Review, 1993, vol. 7, no. 3—4, pp. 227—240. DOI: 10.1007/BF00849556.
13. Bai Y., Zhao J., Wei T. et al. A Survey of Explainable Knowledge Tracing, Applied Intelligence, 2024, vol. 54, pp. 6483—6514. DOI: 10.1007/s10489-024-05509-8.
14. Maloyan N., Ashinov B., Namiot D. Investigating the Vulnerability of LLM-as-a-Judge Architectures to Prompt-Injection Attacks, International Journal of Open Information Technologies, 2025, vol. 13, no. 9. DOI: 10.48550/arXiv.2505.13348.
15. Hwang Y., Lee D., Kang T. et al. When Wording Steers the Evaluation: Framing Bias in LLM Judges, arXiv, 2026. DOI: 10.48550/arXiv.2601.13537.