Mythixs: A Generative AI-Based Mythology Grounded Story Generator using (RAG) - A Review

Abstract

Retrieval-augmented generation RAG serves as an effective system which improves both the factual base and contextual precision of large language models situated within the realm of LLMs. The research study provides an interdisciplinary examination which demonstrates how RAG successfully created mythical stories that meet both ethical standards and cultural norms. The paper analyzes the development and utilization of a Mythology Grounded Story Generator by combining information from fifty research sources which cover three disciplines of global mythology technical RAG architecture and AI ethics. The study examines technological developments which include hybrid retrieval systems and vector databases that use FAISS and embedding models which include Sentence-BERT and MiniLM to achieve semantic alignment between legendary texts and created narratives. The narrative uses elements from Indian mythology Greek mythology Norse mythology Egyptian mythology and Chinese mythology to achieve both authentic storytelling and deep symbolic meaning. Ethical discussions about bias and fairness and transparency in narrative creation receive guidance from international standards which define responsible AI practices. RAG-driven systems serve as a bridge between data-driven creativity and cultural heritage preservation according to the assessment which establishes a foundation for further investigation into narrative AI and digital humanities and ethical storytelling.

Introduction

This paper discusses the use of Artificial Intelligence (AI) and Machine Learning (ML) to improve phishing detection in modern cybersecurity. As phishing attacks become more sophisticated through AI-generated content, deepfakes, and advanced social engineering, traditional detection methods such as blacklists and heuristic filters are no longer sufficient. AI/ML-based systems can analyze emails, URLs, websites, user behavior, and system anomalies in real time to detect phishing attempts with greater accuracy using techniques such as Natural Language Processing (NLP), deep learning, and ensemble learning. However, these systems still face challenges including evolving attack strategies, the need for continuous model updates, and improving model interpretability.

The literature review highlights major advances in Transformer-based NLP models, including BERT, GPT, T5, BART, SBERT, FAISS, and Retrieval-Augmented Generation (RAG). These models improve contextual understanding, semantic retrieval, and content generation, enabling more intelligent and context-aware AI applications. Additional technologies such as Named Entity Recognition (NER), Word2Vec, GloVe, DistilBERT, and multimodal AI models further enhance text processing, retrieval, and generation capabilities.

The paper also presents the NexaLearn framework, an AI-powered educational platform that integrates RAG, BERT, GPT, T5, CNNs, Vision Transformers (ViT), Whisper, FAISS, and LangChain to provide personalized learning, intelligent recommendations, speech-to-text processing, image analysis, and chatbot support. The framework processes multimodal educational data (text, images, audio) and uses semantic retrieval to generate context-aware responses.

Performance comparisons show that NexaLearn outperforms existing transformer-based models, achieving 94–98% accuracy, compared to BERT (80–88%), GPT-2 (82–90%), T5 (85–92%), and RAG (88–94%). Its strengths include high contextual awareness, adaptability, multimodal integration, and personalized learning, although challenges such as high computational requirements, dataset bias, limited interpretability, and long-context processing remain.

Conclusion

The research study investigated how transformer-based architectures and retrieval-augmented generation systems and multimodal learning techniques work together to create educational systems that adapt to student needs and enable interactive learning. Research shows that self-attention mechanisms together with encoder-decoder structures and large language models (LLMs) enable effective creation of question-answer pairs and summarization and transcript comprehension. NexaLearn uses these technological improvements to create a single platform which converts video lectures into customized learning paths through its automatic flashcard system and game-based dashboard and smart chatbot functions. NexaLearn combines cognitive and pedagogical principles with advanced AI technology to solve major problems that affect student engagement and understanding and system scalability. The next phase of research will develop AI-based educational systems through improved model interpretability and better system performance and increased coverage of educational domains.

References

[1] Y. Bengio, A. Courville et al., “Representation learning: A review and new perspectives,” IEEE Trans. Pattern Anal. Mach. Intell., 2013. [2] G. Bradski, “The OpenCV library,” Dr. Dobb’s J. Softw. Tools, 2000. [3] T. Brown, B. Mann et al., “Language models are few-shot learners (GPT-3),” Advances in Neural Information Processing Systems (NeurIPS), 2020. [4] A. Bunt, G. Carenini et al., “Adaptive content presentation for video-based learning systems,” User Modeling and User-Adapted Interaction, 2007. [5] N. Cepeda, H. Pashler et al., “Distributed practice in verbal recall tasks: A review and quantitative synthesis,” Psychological Bulletin, 2006. [6] D. Chen, D. Goldstein et al., “Automatic summarization of lecture videos for quick revision,” IEEE ED-Media Proc., 2018. [7] D. Chen and W. Hersh, “Automatic slide understanding: OCR, diagram parsing and metadata extraction for lecture slides,” IEEE Workshops on Educational Data Mining, 2015. [8] J. Chen and X. Wang, “Diagram parsing in education: State of the art and future directions,” Educ. Technol. Res. & Develop., 2022. [9] M. Chen, J. Tworek et al., “Evaluating large language models trained on code (Codex),” arXiv preprint, 2021. [10] X. Chen, B. Price et al., “Evaluating the accuracy and robustness of LLM-generated answers,” ACL Workshop Papers, 2021. [11] X. Chen and J. Wang, “Automatic diagram recognition in lecture slides using computer vision,” IEEE Trans. Multimedia, 2020. [12] C. Chiarcos, S. Hellmann et al., “Language resources and tools for educational NLP,” LREC Workshops, 2013. [13] S. Deterding, D. Dixon et al., “From game design elements to gamefulness: Defining gamification,” MindTrek Conf., ACM, 2011. [14] J. Devlin, M.-W. Chang et al., “BERT: Pre-training of deep bidirectional transformers for language understanding,” NAACL / arXiv, 2019. [15] S. Ghosal and S. Biswas, “Diagram identification and extraction from lecture frames using OpenCV and Tesseract,” IEEE Conf. Proc., 2018. [16] D. Gibbon, S. Jagadeesan et al., “Evaluating LLMs for educational question generation,” ACL / EMNLP Workshops, 2021. [17] A. Graves, G. Hinton et al., “Speech recognition with recurrent neural networks,” ICASSP / IEEE, 2013. [18] I. Guyon and A. Elisseeff, “An introduction to variable and feature selection,” J. Mach. Learn. Res., 2003. [19] J. Hamari, J. Koivisto et al., “Does gamification work?—A literature review of empirical studies on gamification,” HICSS / IEEE, 2014. [20] G. Huang, Z. Liu et al., “Densely connected convolutional networks (DenseNet),” CVPR, 2017. [21] J. Johnson, M. Douze et al., “Billion-scale similarity search with GPUs (FAISS),” arXiv preprint, 2017. [22] B. Jansen, M. Zhang et al., “Twitter power: Tweets as electronic word of mouth,” J. Amer. Soc. Inf. Sci. & Technol., 2009. [23] S. Kalyuga, “Cognitive load theory: How many types of load does it really need?” Educ. Psychol. Rev., 2019. [24] V. Karpukhin, B. Oguz et al., “Dense passage retrieval for open-domain question answering,” ACL / arXiv, 2020. [25] A. Khosla, T. Zhou et al., “Undoing the damage of unintentional dataset bias,” ECCV / CVPR, 2012. [26] D. Koller and A. Ng, “Machine learning for education: Personalized learning and MOOCs,” Commun. ACM, 2012. [27] D. Koller, A. Ng et al., “MOOCs and open online courses: Research on learning and engagement,” ACM Comput. Surveys, 2015. [28] D. Koller and M. Sahami, “Toward optimal feature selection,” ICML Workshop, 1996. [29] Y. Koren, R. Bell et al., “Matrix factorization techniques for recommender systems,” IEEE Computer, 2009. [30] T. Kulesza, M. Burnett et al., “Why interpretability matters: Explanatory debugging for interactive machine learning,” CHI / IUI Workshops, 2012. [31] T. Kulesza, M. Burnett et al., “Principles of explanatory debugging to personalize interactive machine learning,” IUI / ACM, 2015. [32] H. Lin and P. Chen, “Enhancing learning with retrieval-based question answering systems,” Computers & Education, 2021. [33] H. Lin, K. Xu et al., “LangChain: Modular framework for LLM-powered applications,” arXiv preprint, 2022. [34] P. Lewis, N. Goyal et al., “Question answering with dense retrieval (DPR),” ACL, 2019. [35] P. Lewis, E. Perez et al., “Retrieval-augmented generation for knowledge-intensive NLP tasks (RAG),” NeurIPS / arXiv, 2020. [36] R. Mayer, Multimedia Learning (2nd ed.), Cambridge Univ. Press, 2009. [37] R. Rajan and P. Kumar, “The role of AI and NLP in next-generation EdTech platforms,” Int. J. Artif. Intell. Educ., 2023. [38] A. Radford, J. Kim et al., “Robust speech recognition via large-scale weak supervision (Whisper),” OpenAI / arXiv / PMLR, 2022. [39] A. Radford, J. Wu et al., “Language models are unsupervised multitask learners (GPT-2),” OpenAI Report, 2019. [40] C. Raffel, N. Shazeer et al., “Exploring the limits of transfer learning with a unified text-to-text transformer (T5),” J. Mach. Learn. Res., 2020. [41] B. Shi, X. Bai et al., “An end-to-end trainable neural network for image-based sequence recognition,” IEEE Trans. Pattern Anal. Mach. Intell., 2017. [42] V. Shute, “Focus on formative feedback,” Rev. Educ. Res., 2008. [43] R. Smith, “An overview of the Tesseract OCR engine,” Int. Conf. Document Analysis and Recognition (ICDAR), 2007. [44] I. Sutskever, J. Martens et al., “On the importance of initialization and momentum in deep learning,” ICML, 2013. [45] V. Vasudevan and S. Das, “Automated generation of flashcards and quizzes from lecture transcripts,” Educ. Technol. & Soc., 2019. [46] V. Vasudevan and S. Narayanan, “Automatic transcript alignment and timestamping for video Q&A,” IEEE Workshops, 2020. [47] A. Vaswani, N. Shazeer et al., “Attention is all you need,” Advances in Neural Information Processing Systems (NeurIPS), 2017. [48] M. Zhu and X. Wang, “Gamified dashboards for student engagement: A review of design principles,” J. Educ. Technol. & Soc., 2022.

Copyright

Copyright © 2026 Yash Kumar Mishra, Utkarsh Upadhyay , Shubham Kumar, Aayush Pratap Singh. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.