“The Talent Gap Is Not India's Real AI Problem” — Abdul Nadeem Mohammed's View

In February 2026, Prime Minister Modi inaugurated the India AI Impact Summit at Bharat Mandapam and declared that AI must be a shared resource for the welfare of humanity, not a medium of power. The IndiaAI Mission had already deployed 38,000 GPUs at subsidised rates. But according to ManpowerGroup's 2026 Talent Shortage Survey, 82% of Indian employers now report difficulty filling roles, with AI skills topping the list for the first time. The engineers India is producing are not the engineers its governance ambitions require.

At the centre of that gap is an Indian-born engineer whose career trajectory is itself the proof of concept for what India's AI ambitions require: Abdul Nadeem Mohammed. He started his career at Cyient in India. Then, he built the compliance engine governing employee securities trading at the world's largest asset manager, overseeing more than $10 trillion in assets, where a single misconfigured rule could create conflicts of interest across thousands of employee trades globally. From there, he spent the past year as the sole developer modernising HHS Worker Connect. It is the only platform connecting nine New York City government agencies for eight million residents, and the sole system through which caseworkers can access consolidated, real-time client data across public assistance, child welfare, homeless services, housing, and public health. There is no backup. If it fails, the city's ability to coordinate services for its most vulnerable populations fails with it. 

Additionally, he has published peer-reviewed research at IEEE on machine learning fairness and cloud security, and served as a jury member for the Cases & Faces International Business Award 2026. Also, he is a Council Member of the Association of Information Technology Experts (AITEX), a selective professional body that admits engineers through a rigorous two-stage credential review and committee vote, and a Fellow of Hackathon Raptors. This competitive engineering community awards fellowships through performance-based selection rather than application.

We spoke with Abdul Nadeem Mohammed because he has already done what India is now trying to do, not in a pilot, not in a presentation, but in production, under pressure, with eight million people depending on the outcome. What he found reveals the question India has not yet asked.

Mohammed, India's AI Impact Summit positioned the country as a global use-case capital, yet warned that without accountability, digital infrastructure risks becoming a tool of exclusion. You have actually built AI into a live government platform serving millions. How far is the Summit conversation from the real engineering problem?

The Summit talks about AI. The real problem is the data underneath it, and that work has barely started. Nine agencies, each keeping separate records, each with its own definition of who a person even is. When you put AI on top of that confusion, it does not fix anything. It makes a confident-sounding guess that may be completely wrong. India's Aadhaar and UPI are world-class. But helping a family access welfare requires going much deeper, including ration cards, housing records, and grievance histories. These systems do not speak to each other. The Summit conversation is about the headline. The hard work is in the basement. 

You integrated AI-powered search into Worker Connect, the sole platform, linking all nine of New York City's health and human services agencies for eight million residents, with no backup system. India's IndiaAI Mission is funding similar integrations nationally. Where does this kind of system break down in production?

The AI search worked. But the hard part was never the AI itself; it was everything sitting between the AI and the actual data. Which records does it see? Who decides that? How old is that information? These are not technical details. These are the questions that determine whether a real family gets the right help or the wrong answer. A caseworker reads an AI summary about a family in crisis. It sounds confident and complete. They act on it. But if that data was outdated or pulled from the wrong source, the AI's confident tone just masked the problem rather than surfacing it.

Your IEEE COINS 2025 paper, “Optimising Fairness in Machine Learning: A Hyperparameter Tuning Approach,” shows that bias enters AI systems long before deployment, at the stage where engineers are simply tuning the system for accuracy. India's AI Ethics Framework emphasises fairness. What is the gap between what the framework says and what needs to happen on the ground?

Governance documents describe fairness as a goal. But engineering requires a specific decision about what fairness means at every single step, and those decisions are invisible in most frameworks. Take training data. Engineers optimise for accuracy on whatever data is available. But government data records who interacted with the system, not who needed to. So the AI learns from a skewed picture and reproduces it at scale.

Consider what this means in practice. A farmer in a rural district was denied a food subsidy because the system was trained predominantly on urban data. A family flagged as a fraud risk because their transaction patterns look unusual to a model that has never seen their circumstances before. These are not AI failures. The AI worked exactly as designed. They are failures of assumptions made long before any auditor looked at the system. India's Ethics Framework is a good start. But describing desired outcomes is not the same as requiring the practices that produce them.

India's financial regulators are now pushing banks and brokerages to deploy AI for compliance and fraud detection. You built the system that governs employee trading at BlackRock, the world's largest asset manager. What does that experience tell you about where India's financial AI deployments are most likely to go wrong?

The most common mistake is treating compliance as something you add after the system is built. At BlackRock, compliance was the foundation. Everything else came after. The system I built blocked employee trades that would create conflicts of interest, from day one, not as an afterthought. Going back to add that later is not a small fix. It is starting over. India's banks are at exactly the stage where this decision gets made, and the window to get it right is now. They are rolling out fraud detection and automated loan decisions. The accountability layer needs to be built alongside, not promised for later. A fraud detection system learns from past data. If certain groups were historically over-scrutinised, the system flags them again, millions of people, no explanation given.

Before BlackRock and New York, you spent three years at Cyient in Bengaluru, delivering over ten projects and maintaining a 95% client satisfaction rate. What did that work teach you, and where did it fall short of preparing you for what came next?

Cyient gave me a real engineering foundation. I learned to work across a full stack, to build things that shipped, to operate within enterprise processes and client commitments. Delivering ten projects with a satisfaction rate is nothing. That discipline matters.

What it did not create was the experience of being the last person accountable. At Cyient, there was always a team, a process, an escalation path. That produces excellent engineers. It does not produce engineers who make final decisions alone when millions of people depend on the outcome. The gap between those two things is larger than it looks from the inside.

Engineering credentials come in many forms; some are earned, some are simply purchased. You hold two that fall clearly in the first category: AITEX Council Member and Hackathon Raptors Fellow. What do those recognitions represent, and what did reaching that level require? 

Both mean something precisely because neither is something you can just sign up for. AITEX Council Membership requires independent credential verification, endorsement by an existing member, and review by an expert committee against at least five defined criteria. These are people who have built serious systems themselves. Hackathon Raptors is different; you do not apply, you perform. Either you solve it under pressure, or you do not. But what connects them is simple: you earn both through evidence, not enthusiasm. And that evidence came from work where the stakes were real. 

You have described sole ownership as the experience that India's IT model does not create. As a jury member at the Cases & Faces International Business Award this April, you evaluated cloud migration projects from teams that either had that ownership culture or did not. What separates the migrations that work from those that simply add new complexity?

The pattern was clear. Successful teams could answer two simple questions: what can fail on its own, and what cannot. Their systems matched their failure model exactly.

Teams that got it wrong answered the first question with a diagram. The second question got silence, or a long list of steps that basically described moving everything together at once. That is the tell. If you cannot update one part of your system without calling three other teams first, you have not separated anything. You have just added complexity on top of the same old problem.

For India's government cloud migrations, the lesson is the same. Moving a system to the cloud is not the same as redesigning it. The accountability gaps, the data silos, and the undocumented rules all migrate with it unless someone deliberately addresses them first.

India is building AI infrastructure for hundreds of millions of users. You have built for millions. What is the lesson that is most likely to be underestimated, and what would you tell a young engineer in India who wants to work on these problems?

The salary gap is real. But it is not the deepest problem. What India's IT model does not structurally create is the experience of being the last person accountable. At Cyient, there was always a team, a process, an escalation path. That produces excellent engineers. It does not produce engineers who make final decisions alone when millions of people depend on the outcome. That experience cannot come from a training programme. It comes from real ownership, complete accountability, and genuine authority. The talent gap is not India's real AI problem. The ownership gap in India's startups is beginning to create that. But government digital programmes need to do the same. That shift matters more than any salary adjustment.

Published on: Monday, July 06, 2026, 02:04 PM IST