Artificial intelligence (AI) applications designed for basic phones and limited-connectivity environments are being deployed across healthcare systems in India, demonstrating how task-specific technology could expand access to medical services without requiring expensive infrastructure at every facility.
Shalina Wadhwani of Wadhwani AI US said the organisation’s tuberculosis screening tools had assessed more than 4.5 million people and had been integrated into India’s national tuberculosis platform. The organisation expects the number of assessments to double in the following year.
She also said its clinical decision-support tools had been integrated into India’s national telemedicine platform and had assisted in more than 300 million patient consultations.
The figures were presented at the 2026 World Bank Group and IMF Annual Meetings in Bangkok, where policymakers and technology companies discussed ways to use what they called “small AI” to improve productivity and create jobs in developing economies.
Wadhwani AI is a nonprofit organisation established in 2018 to develop AI applications for underserved populations, with a focus on India and the Global South. Wadhwani said organisations funded by the initiative had developed more than 25 AI solutions across healthcare, education and agriculture.
The healthcare applications described in the presentation include AI-assisted assessments of newborn measurements, tuberculosis screening based on cough sounds, detection of diabetic retinopathy from retinal images and clinical decision support for doctors treating patients remotely.
In one example, a community health worker can film a newborn using a basic phone, allowing an AI model to estimate measurements including weight, height and head circumference. The information can help identify infants who may require additional care.
Another application screens for possible tuberculosis by analysing cough sounds, potentially reducing the need for patients to travel long distances to reach specialist services for initial screening.
For diabetes patients, AI can examine retinal photographs for signs of diabetic retinopathy, helping healthcare workers identify people who may need referral to a specialist before irreversible damage occurs.
Wadhwani said the systems are designed to perform specific tasks, operate on basic phones and function in environments with limited internet connectivity. She also said they cost a fraction of what more complex, general-purpose models would cost to run.
That cost structure is important for governments seeking to deploy healthcare technology across entire populations rather than confining it to a handful of pilot districts.
However, Wadhwani said developing the AI model accounts for only about 20% of the effort required to scale a solution. The remaining work involves partnerships with governments, integration into existing healthcare workflows, training models on local data and ensuring that public institutions develop the skills to operate the systems.
The approach also seeks to preserve the role of healthcare workers. In the clinical decision-support model, AI assists doctors in assessing possible conditions and treatments, while the doctor retains responsibility for the clinical decision.
World Bank Group President Ajay Banga similarly argued that AI could connect rural healthcare facilities to specialists and improve the effectiveness of nurses, midwives and other frontline workers.
The economic case is that technology can extend the reach of existing health personnel, allowing them to serve more patients and potentially creating additional demand for healthcare services.
Wadhwani said the organisation believes governments, funders and technology developers could use AI to improve the lives of one billion people by 2030.
