Researchers from IIT Patna and Stanford University have developed MedSumm, a multimodal approach that combines Hindi-English codemixed medical queries with visual aids to provide a more comprehensive understanding of a patient’s medical condition. Aman Chadha, a researcher from Stanford University, is leading the team that is building India’s first medical large language model (LLM) for Indic languages, including Hindi. Chadha emphasizes the need for LLMs in the healthcare sector, stating that existing models are not focused on Indic languages.
The research team is utilizing Open Hathi as a base LLM and fine-tuning it on medical data in Indic languages. However, Chadha acknowledges the challenge of incorporating medical jargon into the model. Alongside this work, Chadha is also involved in other research projects, such as multimodal question summarization in healthcare.
Despite the shortage of computational resources in India, the researchers are striving to make datasets and AI models more efficient. Chadha expresses a need to overcome constraints and highlights the importance of imagination in driving innovation. He notes that access to GPUs is limited for researchers, which is why they are trying to optimize their models with the available resources.
Regarding data availability, the team is collecting anonymized data to ensure patient privacy. They are also working on building a dataset in addition to adopting techniques for translating information without compromising data quality.
Chadha emphasizes that India has an abundance of talent but lacks sufficient data and computational resources. He calls for increased government funding to support initiatives in this field.
The researchers have high hopes for the impact of their work, believing that once their models and datasets are available, it will encourage further research and utilization by the community. Open source contributions will play a vital role in the advancement of Indic LLMs.