Forget Propreitary LLMs, Open Networks Can Unlock Health AI's UPI Moment

While healthcare AI is moving from pilots to deployment, adoption remains uneven. According to a report by Bain & Co and HealthQuad, most hospitals are stil testing AI in controlled settings. The challenge, increasingly, is of business transformation, where hospitals and healthcare providers tap LLM providers to embed AI directly into areas such as unifying fragmented clinical records and pharmacy systems. 

Open Healthcare Network (OHCN) is betting on a different architecture: let governments control the data and infrastrucure, while AI models remain interchangeable.

The open-source organisation advocating for free digital public infrastructure in healthcare delivery is building infrastructure that governments can use to run healthcare systems.

The non-profit is developing OHCN CARE as a healthcare platform that can run across different cloud environments and plug into different AI systems. The idea is to keep the underlying healthcare infrastructure stable while allowing governments to decide which AI models they want to use.

“Whichever AI the state chooses, we'll run on it. Whatever language model they want, we'll run on it,” Pinaki K, Director of Resources and Partnerships at OHCN,. tells AIM.

Building the Layer Beneath AI

OHCN CARE is designed to be cloud-agnostic. The organisation says the same principle applies to its AI layer.

“We have never created our own language model. We don't want to get into it as of now,” Pinaki reiterates.

Instead, the digital public health and hospital management platform is designed so that an AI model can be treated as another component of the healthcare system. A government could select the model it wants without rebuilding the underlying healthcare workflows.

Data ownership is another part of the architecture. OHCN says it does not hold the state's healthcare data. Instead, the state creates and controls the instance in which the platform operates. “We don't hold any of the data. The state holds the data,” Pinaki says.

This distinction could become increasingly relevant as governments look to deploy AI in healthcare while retaining greater control over sensitive patient information.

The organisation says its platform can also run on different cloud environments and on-premises infrastructure, allowing deployment choices to remain with the state.

The FHIR Foundation

For this model to work, interoperability matters as much as AI.

OHCN CARE is built around FHIR (Fast Healthcare Interoperability Resources), the healthcare data-exchange standard developed by HL7 (Health Level Seven International). The organisation says it adopted the standard to make CARE interoperable with other healthcare systems and to meet the requirements associated with its Digital Public Good positioning.

“We had to go by the FHIR standards to get the DPG licensing and UN and WHO to take us to different countries,” Pinaki explains. 

FHIR is also central to the World Health Organisation's broader digital-health strategy. WHO's SMART Guidelines programme uses standards-based, machine-readable approaches to help countries develop interoperable digital health systems. The approach is designed to be software-neutral, allowing countries to adapt digital components rather than be locked into one technology stack.

WHO has also moved closer to the open-source infrastructure ecosystem, joining the Open Health Stack Software Foundation in 2026. The initiative focuses on open, standards-based digital health infrastructure and interoperability.

Where it All Began 

OHCN's platform did not begin as an AI project. During the COVID-19 pandemic, the organisation built a War Room system for Kerala to coordinate ICU beds, oxygen cylinders, ventilators, ambulances, and patient movement.

“We started off with the War Room software. The software became an instant hit,” Pinaki recalls.

The system initially operated in Kerala before other states adopted it. OHCN subsequently expanded the platform into tele-ICU infrastructure. It says CARE now supports 206 10-bed ICUs across 10 states.

The organisation's technology has also been used for palliative care in Kerala. Its Kerala Care implementation, according to OHCN, supported 3,41,836 patients in its first nine months and recorded 1.6 million home visits.

The system currently connects around 1,405 hospitals and 1,305 NGOs involved in palliative care.

Instead of replacing the existing system, OHCN works to digitise them. “If you try to digitise what is working in the field, the adoption rate is much higher,” Pinaki observes. 

That approach led to rolling out OHCN CARE first at the panchayat level, then district, and finally statewide.

AI Becomes a Plug-in

The platform's AI components follow the same philosophy. One example is OHCN's AI Scribe, which listens to conversations between doctors or nurses and patients, and uses the conversation to populate relevant fields in a patient's record.

But the system does not automatically finalise the information. “The nurse or the doctor or the paramedic has to approve that this is correct,” the director notes.

The organisation describes this as an auditable workflow wherein AI assists with documentation rather than taking control of the healthcare process.

A clinical AI system can be powerful, but it still needs usable healthcare data to produce meaningful results. If patient information is trapped in disconnected systems, inconsistent formats, or manual records, even a sophisticated model will have difficulty operating at scale.

“What our platform does is capture healthcare data in a structured form,” Pinaki clarifies.

He argues that this infrastructure could ultimately be more important than any individual AI capability.

“So any AI component plugged on top of such beautifully structured data, they can churn out awesome results,” he asserts. “More than AI, it is how we packed up our structured data in the system that matters more.”

AI Stack That Can Outlive Its Models

The model-agnostic approach also raises a question about technology lock-in.

Healthcare infrastructure is expected to remain operational for years, while AI models can change within months. A government system tightly coupled to one AI provider could therefore become difficult to upgrade or replace as the market evolves.

OHCN's proposition is to separate the two. The healthcare infrastructure remains standards-based and interoperable, while the AI layer can evolve.

The organisation describes its longer-term ambition as becoming an “operating system for healthcare”, with different capabilities added as plug-ins.

“We are the platform builders,” Pinaki affirms. “That is our goal for the next two, three years.”