Indian AI startup JiviAI, co-founded by former BharatPe Chief Product Officer Ankur Jain, has made headlines with its AI-based healthcare language model, MedX, which claims to outperform those of tech giants Google and OpenAI. 

JiviAI's MedX is now ranked one on the Open Medical LLM Leaderboard hosted by Hugging Face, surpassing OpenAI’s GPT-4 and Google’s MedPaLM2.

Key Achievements

• Top Ranking: Jivi MedX has secured the top position on the Hugging Face Open Medical LLM Leaderboard.
• High Performance: It achieved an impressive average score of 91.65 across nine benchmark categories, outperforming established models from Google and OpenAI.
• Comprehensive Evaluation: The leaderboard assesses medical-specific language models based on their ability to answer questions from medical exams and research. The evaluations include:

1. US Medical Licensing Examination (USMLE)
2. Indian medical entrance exams (AIIMS and NEET)
3. Detailed assessments in clinical knowledge, medical genetics, and professional medicine

Team Composition

The startup has a lean team of physicians, surgeons, AI engineers, and data scientists dedicated to developing transformative healthcare technology. JiviAI aims to enhance the accessibility, affordability, and quality of healthcare through advanced AI technologies.

Industry Impact

JiviAI's achievement highlights the potential of emerging AI startups to challenge established tech leaders and drive innovation in the healthcare sector. MedX's success demonstrates the effectiveness of purpose-built medical large language models in providing accurate and reliable medical information.

This accomplishment not only underscores the growing capabilities of Indian AI startups but also sets a new benchmark for AI-driven advancements in healthcare.

JIVI utilizes its extensive exclusive medical dataset, which comprises millions of medical research publications, journals, clinical notes, and other sources, to train its Jivi MedX LLM. This dataset is one of the largest in the world. Jivi MedX was trained using Odds Ratio Preference Optimisation (ORPO), which involves fine-tuning the model based on specific instructions.

Source: Jivi MedX

Image source: COPILOT