Leveraging AI in ayurvedic agriculture: A RAG chatbot for comprehensive medicinal plant insights using hybrid deep learning approaches

Abstract

Medicinal plants are offering a lot of potential for treatment of various chronic diseases as well as healing wounds, enhancing healthy living for consumers. The Nepalese and Indian agriculture systems are one of the main areas focusing on medicinal plant cultivation, and the abundant availability of these plants in these regions is driving growth in ayurvedic research. Traditional methods for detecting plants as well as generating insights on them are often inefficient and time-consuming due to the manual research need and expertise required in plant and biological lives. In this paper, we develop an advanced LLM (Large Language Model)-powered approach to reliably identify the available medicinal plants and their profitable insights for farmers. We compare multiple deep learning and transfer learning techniques, employing models such as deep convolutional neural networks and advanced transformer models. By training and testing on the dataset that includes varieties of plant types, we select the efficient model after a detailed analysis of a variety of models, dataset split variations, and hyperparameter tuning. The selected model integrates into a retrieval augmented generation (RAG) application capable of providing various insights on the plant identified. The app supports both Nepali and English languages and integrates explainable AI for explaining medicinal plants, their health benefits, and remedies. Results show that the DeiT model achieves 95.97 % accuracy, VGG16 achieves 90.26 %, and a novel hybridized concept with DeiT + VGG16 achieves an accuracy of 96.75 % on a multi-class dataset. The integrated application explains the beneficial insights to users in English as well as local Nepali language.

Keywords

• Ayurveda science
• DeiT
• Transfer learning
• RAG (retrieval augmented generation), Deep learning