AI firm Fractal has developed India’s first multimodal medical LLM, Vaidya. ai. This general-purpose assistant aims to assist users with diagnosis, treatment, and medical advice.
Vaidya.ai uses open-source base models ranging from 30 to 70 billion parameters. It has been trained on over 650,000 images and 200,000 text inputs, enabling it to understand languages like Hindi, Kannada, Tamil, and Telugu, said co-founder and CEO Srikanth Velamakanni.
“It’s not even in the foreseeable future that we will have a situation where we don’t need doctors, and AI may tell you what to do. AI is far from that. The problem we are solving is very different. It’s about aiding consumers in accessing healthcare information which is more reliable than any other sources of information,” he explained.
Alongside Fractal’s established product, Qure.ai, which detects conditions such as tuberculosis, lung cancer, heart failure, and stroke, Vaidya.ai has the potential to be a unique enterprise solution.
Currently, however, Vaidya.ai remains an AI research project rather than a commercial product, according to Velamakanni.
Regarding the accuracy of Vaidya.ai’s responses, Velamakanni mentioned that evaluations are based on PG NEET exam questions.
“PG NEET is the most competitive medical exam with 200,000 postgraduate doctors take every year. Fractal currently ranks at par with top 500 in that system with 83% accuracy,” he said.
Suraj Amonkar, Chief AI Research and Platforms Officer at Fractal, said, “The development involved training models and frameworks using LLM and VLM architectures and validating the models on 20+ years of PG NEET/ PG Medical examination papers, among other metrics to measure how well the models performed. While it offers valuable insights and information, it always encourages users to consult healthcare professionals for medical advice.”
