From Manuscript to Model: Large Language Models as Tools for Ayurvedic Knowledge Preservation and Access

Abstract

Ayurveda, one of the world's oldest continuously practiced medical systems, encodes centuries of clinical knowledge in Sanskrit and regional-language texts that remain largely inaccessible to modern computational tools. The emergence of large language models (LLMs) has opened new possibilities for digitizing, interpreting, and operationalizing this knowledge base at scale. This review synthesizes current research on LLMs applied to Ayurveda, covering domain-specialized models such as AyurParam and AyurGPT, retrieval-augmented generation, machine-learning approaches to Prakriti classification, and digitization of classical texts. It examines key challenges, including hallucination risk, the linguistic complexity of Sanskrit, data standardization gaps, and unresolved ethical and regulatory questions, and contributes a small original case study comparing a general-purpose LLM's behavior against documented domain-specialized systems. The review concludes that domain-adapted LLMs show measurable gains over general-purpose models on Ayurveda-specific tasks, but large-scale clinical validation and governance frameworks remain necessary before such systems can be safely integrated into practice.