India has service-level skills, but lacks deep-tech talent and products: IAIRO founders

India’s artificial intelligence ambitions are being held back not just by a shortage of talent, but by a deeper structural gap — the lack of deep-tech AI talent and a product-first mindset, according to the founders of the Indian Artificial Intelligence Research Organisation (IAIRO).

They said that while India has a large base of engineers suited for services-led work, it lacks the depth of talent required to build original AI models, intellectual property and deep-tech systems.

IAIRO is a national, non-profit research institution established in early 2026 in GIFT City, Gandhinagar. According to its founding director, Dr Amit Sheth, IAIRO can be considered as "India's ISRO" that aims to create, for AI, what ISRO created for space: a concentrated national capability that compounds talent, IP, and execution over decades.

Among its core goals is the creation of world-class AI talent comparable to leading ecosystems in the US and China, alongside building sovereign AI capabilities for India’s strategic autonomy.

Fortune India sat down with Dr Amit Sheth and Professor Dev Niyogi, the founding members of IAIRO, to understand the opportunities and challenges in India’s AI development, a country that produces a large pool of skilled engineers.

“India is just creating services… We are not a product nation,” said Sheth,  adding that while the country is producing skilled engineers, much of India’s talent is geared towards implementation and services, rather than developing foundational AI technologies and globally competitive products.

“What we are producing is people focused on hands-on skills… that serves IT services companies, but not strong IP or global products,” he said.

From skills to systems: fixing India’s AI execution gap

IAIRO, backed by industry and policy stakeholders, is attempting to address what its founders describe as India’s “lab-to-market gap” — the inability to translate research and prototypes into scalable products.

“We want researchers and engineers in the same place… convert prototypes to products, and bring startups into that ecosystem,” Sheth said.

The model goes beyond conventional incubators or think tanks. Instead, IAIRO is positioning itself as an execution layer, where academic research, engineering capability and startup deployment converge.

First principles: data before AI

For Dev Niyogi, the starting point is not flashy AI models, but foundational infrastructure.

“Without data, you cannot create a data-driven system,” he said, outlining that one of the first products IAIRO is working on is a data integration platform that can unify fragmented datasets across domains.

The idea is to build a scalable layer of APIs and data pipelines on top of which startups and developers can build domain-specific applications.

“It’s about putting data from different formats… into a singular unified system from where intelligence can be extracted,” Niyogi said.

Small models, real problems

In a departure from the global obsession with large language models, IAIRO is focusing on smaller, specialised AI systems.

“They are not like large LLMs… but small models solving very specific problems,” Sheth said, adding that such systems are better aligned with real-world decision-making needs.

The emphasis is on decision support systems — AI that helps governments, cities and industries act, rather than just generate text or predictions.

IAIRO is also working on ‘SAMVIT’, an enterprise AI platform aimed at building a new class of models that are smaller, domain-specific and neurosymbolic — combining data-driven learning with structured knowledge to improve efficiency, safety and explainability.

From floods to pollution: AI as a decision engine

IAIRO’s early use cases reflect this approach.

In urban flooding, the gap is not in predicting rainfall, but in hyperlocal forecasting — identifying where exactly rain will hit, down to a granular level.

“What we need is to go from tens of kilometres to tens of metres… that helps prevent nuisance flooding,” Niyogi said.

Similarly, in tackling air pollution, the focus is on identifying the critical sources driving the bulk of emissions.

In order to explain it further, Niyogi referred to the Pareto Principle which states that 80% of outcomes are driven by 20% of causes..

“80% of the problems are caused by 20% of the sources,” he said, advocating the use of city-scale digital twins — virtual replicas of urban systems that allow policymakers to simulate interventions before implementing them.

Neurosymbolic AI: making AI safer and explainable

At the core of IAIRO’s approach is neurosymbolic AI, which combines data-driven neural networks with structured knowledge systems such as rules, policies and regulations.

“In neurosymbolic AI, you are making a black box a grey box,” Sheth said.

This hybrid approach addresses one of AI’s biggest challenges — hallucinations and lack of explainability — by embedding domain knowledge directly into AI systems.

By incorporating regulatory and ethical constraints into knowledge graphs, AI systems can be guided to avoid harmful or non-compliant outputs, while also improving transparency in decision-making.

The founders have stressed that policy will play a critical role in shaping India’s AI trajectory.

“Ethics and data… that is non-negotiable,” Sheth said, stressing that AI systems must be aligned with societal values and regulatory frameworks.

In effect, IAIRO’s approach indicates a broader shift in thinking — from scaling service-oriented talent to building deep-tech capability and systems that generate intellectual property and globally competitive AI products.