Global AI Impact Commons
AI-Powered Clinical Diagnosis and Healthcare Platform Built on Meta's Llama Foundation Model
India
Healthcare
Implementing Organisation
Jivi AI
India, Haryana, Gurugram
Implementing Point of Contact
Ankur Jain
Co-founder and CEO
Contributor of the Impact Story
Carnegie India
Year of implementation
2024
Problem statement
Access to quality healthcare remains limited globally, with significant barriers including cost, availability of specialists, and geographic constraints. Many patients lack 24/7 access to medical guidance, face long wait times for diagnoses, and struggle to understand complex medical information. Healthcare professionals are overburdened with administrative tasks, reducing time for patient care. There is a critical need to democratize access to accurate medical diagnostics and health guidance while supporting healthcare providers with AI-assisted clinical decision-making tools.
Impact story details
Jivi AI is an Indian healthcare AI startup founded in 2023, headquartered in Gurugram, Haryana, India. Co-founded by Ankur Jain and GV Sanjay Reddy, the company is revolutionizing primary healthcare through generative AI. Jivi has developed Jivi MedX, a purpose-built medical Large Language Model (LLM) that ranks #1 on the Open Medical LLM Leaderboard hosted by Hugging Face, outperforming OpenAI's GPT-4 and Google's Med-PaLM 2. The platform leverages Meta's open-source Llama model as its foundation, fine-tuning it specifically for medical applications. Jivi has achieved 1.2 million app installs across 170+ countries, with a 99.7% user satisfaction rate.
AI Technology Used
Large Language Model fine-tuned from Meta's Llama-3-8B using Supervised Fine-Tuning and Odds Ratio Preference Optimization
Key Outcomes
Efficiency
Productivity, Access
Reach, Accuracy
Quality Improvement, User Experience
Satisfaction, Inclusion
Equity
Narrative Outcome
Jivi AI provides AI-powered clinical diagnosis, offering 24/7 access to medical guidance where cost, specialist availability, and geography would otherwise create barriers. The platform generates diagnostic reports in seconds with high-levels of accuracy and user satisfaction. The system demonstrates how well-trained medical AI can extend diagnostic capacity without adding to the administrative burden on healthcare professionals.
Impact Metrics
The global medical AI benchmark performance of Jivi AI's model
Baseline Value
Benchmark scores of Industry-leading models like GPT-4 and Med-PaLM 2 Percentage
Post-Implementation
Jivi AI's model has recorded a 91.65 average score across 9 benchmark categories, including the US Medical Licensing Examination, AIIMS, NEET, clinical knowledge, medical genetics, professional medicine Percentage
The Diagnostic Accuracy of AI Model in Producing Diagnostic Reports
Baseline Value
Standard diagnostic processes
Post-Implementation
Jivi AI's model has recorded a 96.8% accuracy rate in producing diagnostic reports
The Diagnostic Accuracy of AI Model in Producing Diagnostic Reports
Baseline Value
Standard diagnostic processes
Post-Implementation
Jivi AI's model has recorded a 96.8% accuracy rate in producing diagnostic reports
User satisfaction from using the Jivi AI app
Baseline Value
Not Applicable Percentage
Post-Implementation
a 99.7% user satisfaction rate has been recorded on the Jivi AI app Percentage
Time taken by Jivi AI's Mode to generate diagnostic reports
Baseline Value
Traditional diagnostic process timing Seconds
Post-Implementation
Jivi AI's model takes 2.3 seconds to generate each AI diagnosis report Seconds
Implementation Context
Global deployment across 170 countries, with significant user bases in North America, Europe, Asia, and other regions.
People from underserved communities, rural populations, women seeking accessible healthcare, individuals requiring 24/7 medical guidance, and healthcare professionals
Key Partnerships
Andrew Ng's AI Fund, Meta, Hugging Face, NASSCOM, IndiaAI Mission
Replicability & Adaptation
1. Medical datasets should be localized for regional health conditions, drug formularies, and clinical guidelines. 2. Language models can be extended for local languages using similar fine-tuning approaches. 3. Regulatory compliance (HIPAA, GDPR, local medical device regulations) must be addressed per jurisdiction. 4. Integration with local health information systems may require customization. 5. Clinical validation studies are recommended for each new deployment context.
Supporting Materials
* The data presented is self-reported by the respective organisations. Readers should consult the original sources for further details.