AyuRAG: A Retrieval-Augmented Generation Framework for Ayurvedic Knowledge Access
DOI:
https://doi.org/10.63503/acset.978-81-995593-9-4.48Keywords:
Ayurveda, Retrieval-Augmented Generation (RAG), Large Language Models (LLMs), Ontology Alignment, Biomedical NLP, Sanskrit Processing, Traditional Knowledge, Knowledge GraphAbstract
Ayurveda, India’s traditional system of medicine, preserves centuries of clinical knowledge in Sanskrit texts such as Charaka Samhita, Sushruta Samhita, and Ashtanga Hridaya. Although parts of this literature are now accessible thanks to recent digitization efforts, current systems primarily rely on keyword-based search, which restricts the contextual accuracy and semantic depth of clinical queries. This gap could be filled by advances in retrieval-augmented generation (RAG) and large language models (LLMs), enabling multilingual, interpretable, and context-aware access to Ayurvedic knowledge. This paper proposes AyuRAG, a lightweight, domain-adapted LLM framework that integrates RAG with carefully selected Ayurvedic corpora and modern biomedical knowledge bases. AyurRAG guarantees accuracy in retrieval and reasoning, addresses the unique linguistic challenges of Sanskrit, and aligns Ayurvedic disease concepts with contemporary medical ontologies. By using parameter-efficient fine-tuning to achieve scalability for deployment in resource-constrained settings, AyurRAG supports the evidence-based integration of Ayurveda into modern healthcare practices.
References
[1] CSIR/TKDL, “Traditional Knowledge Digital Library (TKDL),” https://www.tkdl.res.in/, 2024, accessed: 2025-08-12.
[2] Government of India Ministry of AYUSH, “AYUSH Research Portal,” https://ayushportal.nic.in/, 2024, accessed: 2025-08-12.
[3] W. Zhang, M. Li, and J. Zhou, “A traditional Chinese medicine clinical knowledge graph integrating disease–symptom–treatment relationships,” Journal of Ethnopharmacology, vol. 295, p. 115356, 2022.
[4] P. Lewis et al., “Retrieval-augmented generation for knowledge-intensive NLP tasks,” NeurIPS, vol. 33, pp. 9459–9474, 2020.
[5] R. Luo et al., “BioGPT: Generative pre-trained transformer for biomed¬ical text generation and mining,” arXiv:2210.10341, 2022.
[6] J. Lee, W. Yoon et al., “BioBERT: a pre-trained biomedical language representation model for biomedical text mining,” Bioinformatics, vol. 36, no. 4, pp. 1234–1240, 2020.
[7] K. V. Sastry, S. Narayanan, and R. Sharma, “Building machine-readable lexicons for Ayurveda: Challenges and opportunities,” Journal of Ayurveda and Integrative Medicine, vol. 10, no. 3, pp. 203–210, 2019.
[8] R. Bhatia and A. Kumar, “Ontology-driven semantic indexing of ayurvedic formulations,” in International Conference on Healthcare Informatics, 2021, pp. 45–54.
[9] G. Huet, “Sanskrit heritage reader,” https://sanskrit.inria.fr/, 2023, online Sanskrit processing toolkit.
[10] M. Prasad, V. Iyer, and A. Sharma, “Ayushehr: Integrating ayurveda with modern electronic health records,” in International Conference on Health Informatics, 2020, pp. 210–218.
[11] Payal Bajaj, Daniel Campos, Nick Craswell, Li Deng, Jianfeng Gao, Xiaodong Liu, Rangan Majumder, Andrew McNamara, Bhaskar Mitra, Tri Nguyen, Mir Rosenberg, Xia Song, Alina Stoica, Saurabh Tiwary, and Tong Wang. MS MARCO: A Human Generated MAchine Reading COmprehension Dataset. arXiv:1611.09268, 2016.
[12] Petr Baudisˇ and Jan Sˇedivy´. Modeling of the Question Answering Task in the YodaQA System. In *CLEF*, pp. 222–228. Springer, 2015.
[13] Jonathan Berant, Andrew Chou, Roy Frostig, and Percy Liang. Semantic Parsing on Freebase from Question-Answer Pairs. In *EMNLP*, pp. 1533–1544, 2013.
[14] Bin Bi, Chenliang Li, Chen Wu, Ming Yan, Wei Wang, Songfang Huang, Fei Huang, and Luo Si. Palm: Pre-training an Autoencoding Autoregressive Language Model for Context-conditioned Generation. In *EMNLP*, pp. 8681–8691, 2020.
[15] Danqi Chen, Adam Fisch, Jason Weston, and Antoine Bordes. Read¬ing Wikipedia to Answer Open-Domain Questions. In *ACL*, pp. 1870–1879, 2017.
[16] Angela Fan, Mike Lewis, and Yann Dauphin. Hierarchical Neural Story Generation. In *ACL*, pp. 889–898, 2018.
[17] Angela Fan, Yacine Jernite, Ethan Perez, David Grangier, Jason Weston, and Michael Auli. ELI5: Long Form Question Answering. In *ACL*, pp. 3558–3567, 2019.
[18] Angela Fan, Claire Gardent, Chloe Braud, and Antoine Bordes. Aug¬menting Transformers with KNN-Based Composite Memory for Dialog. *TACL*, 8:82–98, 2020.
[19] Thibault Fe´vry, Livio Baldini Soares, Nicholas FitzGerald, Eunsol Choi, and Tom Kwiatkowski. Entities as Experts: Sparse Memory Access with Entity Supervision. In *EMNLP*, pp. 4937–4951, 2020.
[20] Marjan Ghazvininejad, Chris Brockett, Ming-Wei Chang, Bill Dolan, Jianfeng Gao, Wen-tau Yih, and Michel Galley. A Knowledge-Grounded Neural Conversation Model. In *AAAI*, 2018.
[21] Kenton Lee, Ming-Wei Chang, and Kristina Toutanova. Latent retrieval for weakly supervised open domain question answering. In *ACL*, pp. 6086–6096, Florence, Italy, July 2019. doi:10.18653/v1/P19-1612. URL: https://www.aclweb.org/anthology/P19-1612.
[22] Mike Lewis et al. BART: Denoising sequence-to-sequence pre-training for natural language generation, translation, and comprehension. arXiv preprint arXiv:1910.13461, 2019. URL: https://arxiv.org/abs/1910.13461.
[23] Jiwei Li et al. A diversity-promoting objective function for neu¬ral conversation models. In *NAACL-HLT*, pp. 110–119, San Diego, California, June 2016. doi:10.18653/v1/N16-1014. URL: https://www.aclweb.org/anthology/N16-1014.
[24] Margaret Li, Jason Weston, and Stephen Roller. Acute-eval: Improved dialogue evaluation with optimized questions and multi-turn compar¬isons. arXiv:1909.03087, 2019. URL: https://arxiv.org/abs/1909.03087.
[25] Hairong Liu et al. Robust neural machine translation with joint textual and phonetic embedding. In *ACL*, pp. 3044–3049, Florence, Italy, July 2019. doi:10.18653/v1/P19-1291. URL: https://www.aclweb.org/anthology/P19-1291.
