AyurGPT: Fine-Tuning a Medical LLM for Multilingual Ayurveda Consultations

This talk shares key findings, learnings, and a demo from developing AyurGPT, a large language model specialized for medicine consultations on Ayurveda, the traditional Hindu system of medicine. This LLM is optimized to pursue these goals: Enhance understanding of Ayurvedic terminology Improve diagnostic approach comprehension Better treatment recommendation capabilities AyurGPT is a fine-tuned version of BioGPT-Large-PubMedQA, built using the PubMedQA dataset for synthetic data generation, and a mix of real and synthetic Ayurveda-specific consultations. The model supports English, Kannada, Tamil, and Malayalam, and we’ll share metrics on its performance in the following tasks: Medical question answering Generation of culturally appropriate responses in multiple languages