Why I Chose to Build AI Products for India First

The Assumption Baked Into Most AI Products

Most AI products are built assuming: reliable broadband connectivity, users who read English fluently, relatively modern hardware, and payment infrastructure that supports subscription billing. These assumptions describe the median Silicon Valley user fairly well. They describe the median Indian user poorly.

India has roughly 1.4 billion people, 22 languages in the Eighth Schedule to the Constitution [1], household incomes far below the assumptions baked into most pricing, and significant connectivity variation — from fiber-connected urban apartments to 2G edge connectivity in rural areas. Building for India means building for a user whose needs and constraints differ fundamentally from the assumptions baked into most AI design.

I chose to build for this user deliberately. Here's why I think it makes our products stronger.

Constraints Produce Better Engineering

When you can't assume fast internet, you build systems that work offline or on low bandwidth. SmartON runs entirely on-device because rural Ahmedabad users in markets and on buses don't have reliable connectivity. This constraint produced a product that also works in basements, tunnels, and any connectivity-poor environment globally — not just India.

When you can't assume a modern device, you optimize. SmartON runs YOLOv8n with TensorRT quantization because it needs to run on a ₹8,000 Jetson Nano with 4GB RAM. The same optimization makes it faster and cheaper to run on higher-end hardware. Constraints force optimization that benefits everyone.

When you can't assume English fluency, you build multilingual from the start. MIRA handles Gujarati, Hindi, and English because those are the languages our users speak. The multilingual architecture we built for India doesn't need to be retrofitted when expanding to other language markets — it's foundational.

The Market Opportunity Is Genuinely Large

Indian AI adoption is accelerating. Resist any single number for its size. Published estimates run from IDC's roughly $6 billion of AI spending by 2027 to a NASSCOM-BCG projection of $17 billion by 2027 at 25-35% CAGR, to a Wheebox forecast the government itself cites of $28.8 billion by 2025 at 45% CAGR [2]. A four-to-fivefold spread is not measurement error — it is four different definitions of what counts as "the AI market". Use the direction, which every estimate agrees on, and distrust the decimals. Sectors that are particularly active: agriculture, healthcare, financial services, education, and government services.

The opportunity is not uniform. Enterprise AI for large Indian corporations is a well-served market — Bangalore-based product companies, multinationals, and SaaS vendors are all competing here. The underserved opportunity is AI for the "missing middle" — businesses and individuals with real needs but not the budget or infrastructure assumptions of enterprise software. This is where Sunbots operates with our management platform, and where SmartON operates in assistive tech.

Building for this market requires accepting lower per-unit economics but serves a significantly larger addressable population than enterprise-only approaches.

The Cultural Advantage of Building at Home

I grew up in Ahmedabad. I know what a Gujarati market sounds like, what lighting conditions look like in a typical Indian shop, and how visually impaired users in Indian cities navigate their environment differently from visually impaired users in European cities. This knowledge is not documentable — it's accumulated from experience and cannot be bought with research budgets.

When we test SmartON with users in Ahmedabad, I'm not an outsider studying a foreign market. I'm an insider who can distinguish between problems that are universal and problems that are specific to this context. That distinction matters for product decisions.

The companies that will dominate Indian AI are likely to be built by people who understand Indian contexts from the inside. The head start from building here first is real.

The Global Applicability Surprise

The assumption when building for India is that you're limiting your market. The experience has been the opposite. A product that works without internet works anywhere with poor connectivity — developing markets globally, rural areas in wealthy countries, and constrained environments like hospitals and factories. A product that supports code-switching between multiple languages is architecturally better equipped to handle any multilingual market.

India-first constraints haven't limited our market. They've produced product capabilities that differentiate us in markets we weren't originally targeting.