Ankur Jain , the former Chief Product Officer of BharatPe, ventured into the fields of AI and healthcare after leaving the fintech giant last year. Together with G V Sanjay Reddy, Chairman of Reddy Ventures, Jain co-founded Jivi.ai , marking his second entrepreneurial venture. Jain’s background in AI and machine learning has fueled the creation of a pioneering startup aimed at transforming primary healthcare on a global scale.
Founding Jivi.ai
Jivi.ai was established in January 2024 by Ankur Jain. Within a few months, the startup introduced its initial model, Jivi MedX, which surpassed well-known models such as Google’s Med-PaLM 2 and OpenAI’s GPT-4 on the Open Medical LLM Leaderboard. This achievement set a high benchmark for the company, signalling its potential to revolutionise the healthcare industry.
Jivi MedX and future models
Jivi MedX is just the beginning of a series of healthcare-focused models that the company plans to release. Jain disclosed plans for a "model cluster," which includes specialised models for various medical fields such as diabetes and ophthalmology. The next model in line is a vision model, part of a multimodal MedX suite.
Ankur Jain's vision for global health
Jain’s ambition with Jivi.ai is to develop an AI medical companion accessible to the entire global population, free of charge. While AI is making strides in areas like drug discovery and genomics, Jivi.ai focuses on primary healthcare, aiming to provide comprehensive support to both end users and medical professionals.
Jivi.ai is beyond a chatbot
Jivi.ai’s AI medical companion is designed to be more than just a chatbot. It will incorporate voice capabilities, allowing users to communicate with the model in their native languages. The goal is to support the top 25% of world languages, covering around 90% of the global population. Jain anticipates that voice capabilities will become mainstream within the next six months, enhancing the model's usability.
Navigating regulatory challenges
Healthcare is one of the most regulated industries, and Jain is fully aware of the regulatory challenges that come with incorporating AI. Jivi.ai adheres to standards like ISO 42,001 and complies with evolving regulations both in India and globally. Jain believes that regulatory frameworks will ensure only serious players enter the market, ultimately benefiting consumers with high-quality products.
Jivi MedX powered by Llama 3
Jivi MedX leverages the Llama 3 model, with parameters ranging from 8 billion to 70 billion. The startup initially tested existing models like Mistral and Llama 2, but none met the accuracy requirements for medical applications. After extensive experimentation, Jivi.ai settled on odds ratio preference optimization (ORPO), conducting around 150 experiments to achieve the desired accuracy. Currently, Jivi MedX boasts an average score of 91.65% across nine benchmark categories, with ongoing efforts to reach 95%.
Team and development process
Despite being an AI startup, Jain’s first hire was a doctor. Nearly 40% of Jivi.ai’s 24-member team comprises medical professionals. This collaborative approach ensures that the models are rigorously evaluated and refined based on feedback from the medical community. Jain emphasised the importance of involving doctors from the outset to create a reliable and effective product.
Data and continuous improvement
Jivi.ai trains its models using domain-specific data sourced from web scraping, medical journals, research papers, and clinical notes. This comprehensive knowledge base is continually updated to incorporate new developments in medical research. The company aims to ingest new data every week to keep the models current and accurate.
Monetization strategy
While creating an advanced AI model is a significant achievement, monetizing it presents its own challenges. Jivi.ai plans to offer its product to general consumers for free while developing a B2B segment. Hospitals, pharmaceutical companies, and insurance providers can integrate Jivi’s product into their services as a software-as-a-service (SaaS) offering. Additionally, Jivi.ai plans to generate revenue through premium services and cross-selling opportunities, such as connecting users to diagnostic services and selling medical devices.
Also read | eSIM vs physical SIM: Making the best choice for your needs
Founding Jivi.ai
Jivi.ai was established in January 2024 by Ankur Jain. Within a few months, the startup introduced its initial model, Jivi MedX, which surpassed well-known models such as Google’s Med-PaLM 2 and OpenAI’s GPT-4 on the Open Medical LLM Leaderboard. This achievement set a high benchmark for the company, signalling its potential to revolutionise the healthcare industry.
Jivi MedX and future models
Jivi MedX is just the beginning of a series of healthcare-focused models that the company plans to release. Jain disclosed plans for a "model cluster," which includes specialised models for various medical fields such as diabetes and ophthalmology. The next model in line is a vision model, part of a multimodal MedX suite.
Ankur Jain's vision for global health
Jain’s ambition with Jivi.ai is to develop an AI medical companion accessible to the entire global population, free of charge. While AI is making strides in areas like drug discovery and genomics, Jivi.ai focuses on primary healthcare, aiming to provide comprehensive support to both end users and medical professionals.
Jivi.ai is beyond a chatbot
Jivi.ai’s AI medical companion is designed to be more than just a chatbot. It will incorporate voice capabilities, allowing users to communicate with the model in their native languages. The goal is to support the top 25% of world languages, covering around 90% of the global population. Jain anticipates that voice capabilities will become mainstream within the next six months, enhancing the model's usability.
Navigating regulatory challenges
Healthcare is one of the most regulated industries, and Jain is fully aware of the regulatory challenges that come with incorporating AI. Jivi.ai adheres to standards like ISO 42,001 and complies with evolving regulations both in India and globally. Jain believes that regulatory frameworks will ensure only serious players enter the market, ultimately benefiting consumers with high-quality products.
Jivi MedX powered by Llama 3
Jivi MedX leverages the Llama 3 model, with parameters ranging from 8 billion to 70 billion. The startup initially tested existing models like Mistral and Llama 2, but none met the accuracy requirements for medical applications. After extensive experimentation, Jivi.ai settled on odds ratio preference optimization (ORPO), conducting around 150 experiments to achieve the desired accuracy. Currently, Jivi MedX boasts an average score of 91.65% across nine benchmark categories, with ongoing efforts to reach 95%.
Team and development process
Despite being an AI startup, Jain’s first hire was a doctor. Nearly 40% of Jivi.ai’s 24-member team comprises medical professionals. This collaborative approach ensures that the models are rigorously evaluated and refined based on feedback from the medical community. Jain emphasised the importance of involving doctors from the outset to create a reliable and effective product.
Data and continuous improvement
Jivi.ai trains its models using domain-specific data sourced from web scraping, medical journals, research papers, and clinical notes. This comprehensive knowledge base is continually updated to incorporate new developments in medical research. The company aims to ingest new data every week to keep the models current and accurate.
Monetization strategy
While creating an advanced AI model is a significant achievement, monetizing it presents its own challenges. Jivi.ai plans to offer its product to general consumers for free while developing a B2B segment. Hospitals, pharmaceutical companies, and insurance providers can integrate Jivi’s product into their services as a software-as-a-service (SaaS) offering. Additionally, Jivi.ai plans to generate revenue through premium services and cross-selling opportunities, such as connecting users to diagnostic services and selling medical devices.
Also read | eSIM vs physical SIM: Making the best choice for your needs
You may also like
Suella Braverman says 'sorry' after winning huge majority as she eyes up leadership bid
Jeremy Corbyn's stinging three-word verdict on whether Keir Starmer will make a good PM
Roses will flower longer and grow stronger when fed 3 kitchen scraps this July
Chelsea get £50m transfer green light by Enzo Maresca target as £17m wonderkid loan already confirmed
