On a sweltering afternoon in Mumbai, a delivery rider struggled to follow English instructions on his app, losing time and pay with every mistake. Existing translation services were buggy and cumbersome. Until one day, when suddenly, the app “spoke” to him in Marathi. What seems like a small change is in fact the front line of a larger battle. This is courtesy of a new generation of Indian-language AI tools, trying to ensure that critical services like banking, healthcare, education, and government are not out of reach to millions of Indians because of the divide between the digital world and their respective mother tongues.
These efforts are being driven by the Modi government as part of the next phase of India's digital public infrastructure, with the explicit goal of leveraging technology to reduce inequalities and expand access. Just as Aadhaar gave Indians a digital identity and UPI made digital payments universal, projects like Bhashini and BharatGPT, flagship components of the broader IndiaAI strategy, aim to provide every citizen with a unified digital voice, regardless of the language they speak.
Bhashini is the foundation. Launched in 2022, it aims to make digital content and services available in all major Indian languages through cutting-edge AI. Today, it processes over ten million translation requests daily. Crucially, it is not limited to dominant languages, thanks in part to the Bhasha Daan initiative, which has built datasets in historically neglected tongues like Khasi, Manipuri, and Bodo. Its multilingual interfaces now run across government portals from e‑learning to telehealth, bringing services once possible only in English into vernacular heartlands.
Running on this foundation is BharatGPT. Built with Indian institutes and startups, it is slated to be India's first sovereign large language model. Unlike global systems trained predominantly on English, BharatGPT is tuned to Indic datasets, making it context-aware and culturally grounded. It's also multimodal: supporting text, voice, and video dialogues, and designed explicitly for public use cases. It already supports citizen services from ticketing to insurance helplines.
These efforts are already having a transformational impact on ordinary citizens. Take rural women entrepreneurs. With AI-powered translation and voice assistance, they can now access government credit schemes without needing intermediaries. Farmers can query weather or mandi prices in their dialect. Patients can talk to telemedicine bots and receive credible advice.
But this is not a story of the state alone. The ecosystem matters as much as the platform. AI4Bharat, housed at IIT Madras, is building the open corpora that cover India's less-spoken languages. Sarvam AI is releasing conversational models as part of the IndiaAI mission. International partners are part of the mix, but under Indian terms. Google Cloud hosts BharatGPT with data sovereignty rules. NVIDIA has gone so far as to highlight BharatGPT as proof that small, efficient models can be more culturally relevant than trillion-parameter behemoths. This fusion of state vision, academic research, entrepreneurial drive, and localisation is exactly the formula that made UPI a global lesson in digital payments, and it is at play here once more.
The road ahead will depend on whether India can deepen these foundations, though. The quality of data and coverage across dialects must expand so that no community is left out. Access to computing cannot remain the preserve of big players; smaller universities and firms need a fair share and the physical infrastructure to make use of it. Public services must embed AI in the front lines, like schools or clinics, where inclusion matters most. Above all, India must invest as much in people as in platforms. The country's AI talent is still heavily concentrated at the low- and mid-tier, and the pool of cutting-edge researchers remains small and prone to migrating overseas. Without building and retaining a cadre of top-tier AI scientists, India risks becoming a consumer rather than a pioneer. Complementing this, AI literacy in Indic languages for ordinary users will make way for a much more meaningful, secure, and sustainable path toward their digital empowerment.
Placed globally, India's approach is unusual. In the US, the AI race is corporate-led, with firms competing to build increasingly larger models in search of a commercial edge. Multilingual adaptation or customisation to run on low-capacity systems is usually an afterthought. In China, the state funds sovereign AI projects, but linguistic diversity is not the goal; instead, these systems aim to create the conditions for monolingual homogeneity, as this would enable tighter optimisation and better state control over information. The European Union's own efforts are fragmentary and lack a public-first LLM, meaning that they are slow to be deployed in the real world. By contrast, India's model is pragmatic and, more importantly, is already in action. It does not chase size, nor centralise control, but steers AI to solve a development problem on how to give all citizens a voice online.
The implications of that go well beyond India's borders. Much of the Global South faces the same tension between linguistic diversity and digital access. If India can demonstrate through this approach that billions can seamlessly integrate into the digital economy, all while continuing to retain their distinctive linguistic identities, the impact could be transformative, creating a model that could shape not just its own future, but the world's.
(The author is a graduate from the Harvard Kennedy School and a public policy consultant)
Disclaimer: These are the personal opinions of the author
