FlashSnap: AI-Powered Flashcard Generation and Spaced Repetition via Large Language Models

This paper has presented FlashSnap, an AI-powered flash-card generation and spaced repetition platform that addresses the two primary barriers to effective flashcard-based study: the high authoring cost of quality decks and the absence of intelligent review scheduling. By integrating a multi-provider LLM cascade with SM-2 spaced repetition scheduling within a production-grade serverless web architecture, FlashSnap reduces deck creation time by an estimated 95 percent relative to manual authoring and provides adaptive review scheduling that is projected to reduce total review effort by approximately 40 percent for equivalent retention outcomes. The system further provides a Forgetting Curve Visualiser for learner metacognition, a community marketplace for deck sharing, and open-format data export for portability. The platform is deployed and accessible, demonstrating that research-grade memory science can be operationalised within a modern web application at negligible marginal cost to the end user. The authors anticipate that the architectural patterns and prompt-engineering methodology described herein will be of value to researchers and practitioners developing intelligent tutoring and learning management systems.

[1]     H. P. Bahrick and E. Phelps, “Retention of Spanish vocabulary over 8 years,” J. Exp. Psychol. Learn. Mem. Cogn., vol. 13, no. 2, pp. 344–349, 1987. [2]     J. D. Karpicke and J. R. Blunt, “Retrieval practice produces more learning than elaborative studying with concept mapping,” Science, vol. 331, no. 6018, pp. 772–775, 2011. [3]     D. Elmes, “Anki — powerful, intelligent flashcards,” 2006. [Online]. Available: https://apps.ankiweb.net [4]     P. A. Wozniak and E. J. Gorzelanczyk, “Optimization of repetition scheduling with application to learning of foreign-language vocabulary,” Acta Neurobiol. Exp., vol. 54, no. 1, pp. 59–62, 1994. [5]     H. Ebbinghaus, Memory: A Contribution to Experimental Psychology. New York: Teachers College, Columbia University, 1885. [Online]. Available: https://psychclassics.yorku.ca/Ebbinghaus/memory.htm [6]     S. Leitner, So lernt man lernen. Freiburg: Herder, 1972. [7]     M. Heilman and N. A. Smith, “Good question! Statistical ranking for question generation,” in Proc. NAACL HLT, 2010, pp. 609–617. [Online]. Available: https://aclanthology.org/N10-1096/ [8]     J. Wang et al., “Generalizing to unseen domains: A survey on domain generalization,” IEEE Trans. Knowl. Data Eng., vol. 35, no. 5, pp. 4978–4997, 2023. [9]     Quizlet Inc., “Quizlet — study tools and flashcards,” 2005. [Online]. Available: https://quizlet.com [10]     R. C. Atkinson, “Optimizing the learning of a second-language vocab-ulary,” J. Exp. Psychol., vol. 96, no. 1, pp. 124–129, 1972. [11]     T. Brown et al., “Language models are few-shot learners,” in Proc. NeurIPS, vol. 33, 2020, pp. 1877–1901. [Online]. Available: https: //arxiv.org/abs/2005.14165 [12]     OpenAI, “GPT-4 technical report,” 2023. [Online]. Available: https:// arxiv.org/abs/2303.08774 [13]     Google, “Gemini: A family of highly capable multimodal models,” 2023. [Online]. Available: https://arxiv.org/abs/2312.11805 [14]     P. Lewis et al., “Retrieval-augmented generation for knowledge-intensive NLP tasks,” in Proc. NeurIPS, 2020. [Online]. Available: https://arxiv. org/abs/2005.11401 [15]     S. Settles and B. Meeder, “A trainable spaced repetition model for language learning,” in Proc. ACL, 2016. [Online]. Available: https: //aclanthology.org/P16-1174/