AyuRAG: Ayurvedic Knowledge Integration with Retrieval-Augmented Lightweight Large Language Models

Ayurveda, a traditional system of medicine, is boundless and rich in historical connotations, but it often remains unutilized because of fragmented data sources as well as the absence of access to properly structured, substantial literature. Healthcare professionals face difficulties in accessing and integrating Ayurvedic knowledge, hindering their ability to derive actionable understandings from comprehensive literature. Traditional methods are not adequate to provide structured, accessible databases, which limits the utility of Ayurveda in modern healthcare. To bridge this gap, a Large Language Model (LLM) with Retrieval-Augmented Generation (RAG) is proposed in this work. This system enables concise, context-dependent retrieval of Ayurvedic knowledge and uses RAG to dynamically retrieve relevant chunks of information from the repository based on the query and uses LLMs to generate contextually accurate responses. This hybrid response accounts for highly specific and relevant responses. By utilizing Ayurvedic data from unstructured, publicly available sources, this system is structured around key medical domains, focusing on diseases and their cures as per Ayurveda. Additionally, various techniques were used for faster inference and mapping for more accurate and efficient workflow. This approach overcomes shortcomings like fragmentation and inaccessibility of Ayurvedic data. The scalability of this approach provides future expansion, offering a versatile platform for integrating Ayurvedic principles into modern medical practices. In future expansions a wider dataset with a focus on other aspects of Ayurveda could be included, extending the capabilities to turn it into a more specialized Ayurvedic care for individuals.