Image credit: Neuralix AI — official company visual.
The Indian Army’s artificial intelligence cannot live on the public internet. It cannot phone home to a foreign cloud, leak its training data through a third-party API, or depend on model weights owned by a commercial vendor in another country. That constraint — absolute data sovereignty — is what makes most AI development playbooks useless in a defence network. And it is exactly the problem Neuralix AI, a Bengaluru deep-tech startup, says it has solved.
The company has built EKAM, a unified sovereign military generative-AI platform, and a military-specific AI foundation model trained from the ground up — both running entirely inside the Indian Army’s secure, air-gapped digital environment. EKAM was launched by the Hon’ble Raksha Mantri Shri Rajnath Singh at the Chanakya Defence Dialogue, under the Ministry of Defence’s iDEX ADITI 2.0 programme.
The platform: EKAM
EKAM is described by the company as India’s first indigenous defence AI-as-a-Service platform. It is a unified environment designed to host multiple AI models, specialised services and military use cases — all operating within the Army’s own network and compute, isolated from the public internet and external cloud services.
According to Neuralix, the platform supports the ground-up military foundation model as well as smaller, specialised language models tuned for specific Army requirements. The company’s website lists nine operating domains for the platform, spanning defence and public services, with applications including battlefield-intelligence automation and decision support deployable both at headquarters and on edge devices in air-gapped, bandwidth-constrained sectors.
A foundation model built from scratch
Rather than adapting a general-purpose or open model for military use, Neuralix says it developed a foundation model specifically around defence requirements — with its own architecture, its own weights, and a training process designed for the military domain. The model uses a Mixture-of-Experts architecture and, the company says, was developed within the Indian Army environment so sensitive data remained under Army control throughout.
More Read
The harder engineering went into the language itself. “The military writes its own language — the forms, the jargon, the way they interpret results. It’s very different from regular natural language,” Neuralix co-founder and CEO Vikram Jayaram told Analytics India Magazine. The team redesigned transformer architectures, built custom tokenisers for military terminology, and trained the models from scratch — deliberately, to reduce hallucinations and improve domain inference on defence documents, doctrine and strategy.
The model is designed to understand defence terminology, military concepts, doctrine and complex Army documents, generating structured insights for authorised users. Even the knowledge-and-retrieval architecture, the company says, was designed around military information requirements rather than bolting conventional retrieval onto a generic language model.
The engineering problem: building AI with no internet
Creating modern AI inside an air-gapped defence network was, the company admits, a substantial engineering challenge. Contemporary AI development assumes constant access to software repositories, model hubs, external libraries, APIs and cloud compute. Inside the Army’s network, none of those dependencies could simply be fetched on demand.
Neuralix engineers had to securely bring required dependencies into the environment, modify components where necessary and, in some cases, develop or adapt software specifically so the complete AI stack could operate efficiently without external connectivity. The company used what it calls a forward-deployed engineering approach — its technical team worked inside the environment provided by the Army’s Directorate General of Information Systems (DGIS), developing the technology closely around Army requirements. Sensitive data never had to leave the Army environment for the system to be developed, tested and refined. The company says the platform and the foundation model were built in just over a year.
Neuralix emerged as the selected technology developer for the programme — the Indian Army’s AI-as-a-Service with infrastructure-setup challenge under iDEX ADITI 2.0 — following a competitive process involving multiple contenders. DGIS serves as the Army’s steering and user agency; the Defence Innovation Organisation is the contracting agency.
Sovereignty in three layers
The company frames the achievement as sovereignty across three layers. Model sovereignty: the architecture, weights and evolution of the AI remain in controlled custody rather than depending on an external commercial provider. Data sovereignty: confidential Army information stays inside the Army environment. Infrastructure sovereignty: the complete system operates on Army-controlled compute and networks, with no dependence on public cloud infrastructure or external APIs.
The intended applications are broad: command support, document-intelligence analysis, speech interfaces, translation engines, operational analytics and secure data integration — each aimed at improving decision-making, situational awareness and operational efficiency within the armed forces.
The company behind it
Neuralix AI Pvt. Ltd. is a Bengaluru-rooted startup founded by Vikram Jayaram — who spent decades across aerospace, academia and the energy sector — along with co-founder Rohit Gangwal. The company began with industrial AI: physics-grounded models for industrial plants, utilities and critical infrastructure, turning sensor and SCADA data into predictive maintenance, digital twins and anomaly detection. Its defence push began with a no-cost pilot for the Ministry of Defence, which the company says succeeded — leading to the iDEX grant in 2024 and the EKAM programme that followed.
The company was named AI Innovator at The Economic Times’ “One of a Kind Awards”, recognising its application of AI to complex industrial operations across energy, manufacturing and defence.
What Neuralix is doing is not the flashiest use of AI — there are no viral demos here. It is something rarer: a ground-up foundation model, built inside a vault, trained to speak the Army’s language, and launched by the Defence Minister himself. In the global race to put AI inside critical national infrastructure, India now has a working demonstration of the sovereign-by-design approach.
