By the crowded roads of Mumbai and the quiet corridors of government offices, a quiet technological revolution is unfolding—one that seeks to make artificial intelligence speak not just English or Hindi, but the full linguistic diversity of India.
On most days, Vineet Sawant weaves through Mumbai’s relentless traffic on his scooter, a large delivery box strapped to the back. The job is physically demanding and mentally draining, especially in a city where every second counts. But for Sawant, the stress of navigating potholes and impatient drivers was once compounded by something less visible: language.
Sawant’s first language is Marathi. Like millions of Indians, he understands some English but struggles to read it fluently. When he joined Zepto, a fast-growing online grocery delivery company promising lightning-fast service, the language gap quickly became a problem.
“At first, it was difficult,” he recalls. “Everything was in English. I could understand some things, but I wasn’t comfortable. I used to ask other delivery guys what the instructions meant.”
For a company built on speed and efficiency, that friction mattered. So last year, Zepto partnered with Reverie Language Technologies to roll out an AI-powered translation service for its delivery app. Today, drivers can choose from six Indian languages, instantly translating customer instructions into their mother tongue.
The difference, Sawant says, has been transformative.
“I don’t have to guess anymore. If a customer writes ‘ring bell,’ I get it in Marathi. It’s clear. I don’t make mistakes.”
His daily deliveries have tripled—from around 10 parcels a day to nearly 30. More importantly, he says, “It makes us feel like we belong.”
Sawant’s experience captures both the promise and the complexity of one of the most ambitious AI challenges anywhere in the world: making artificial intelligence work seamlessly across 22 official languages—and hundreds of dialects—in a country as vast and diverse as India.
A Linguistic Superpower, and a Digital Challenge
India is often described as a linguistic superpower. The Constitution recognizes 22 official languages, and linguists estimate that hundreds of regional languages and dialects are spoken across the country. This diversity is a cultural asset—but in the digital age, it has also become a fault line.
“Without technology that understands and speaks Indian languages, millions are excluded from the digital revolution,” says Professor Pushpak Bhattacharyya of IIT Mumbai, one of India’s leading experts on AI and language technologies. “This affects education, governance, healthcare, and banking.”
For years, India’s digital services—both public and private—have largely operated in English and, to a lesser extent, Hindi. That has left vast swathes of the population navigating apps, websites, and automated systems in languages they are not fully comfortable with.
The arrival of generative AI systems like ChatGPT has made the gap even more visible. These models promise conversational interfaces that could democratize access to information—but only if they can communicate effectively with users in their own languages.
Why AI Struggles With Many Languages
At its core, modern AI depends on data—vast quantities of it. Language models are trained on web pages, books, transcripts, and other text sources. For English and other globally dominant languages, this data is abundant and relatively easy to access.
For many Indian languages, the situation is starkly different.
“The main challenge is not data in general, but high-quality data,” Professor Bhattacharyya explains. “Coarse data exists, but it needs heavy filtering. For many regional and tribal languages, the data is not digitized at all.”
This imbalance means that AI systems trained primarily on English or other major languages struggle with nuance, grammar, idioms, and cultural context when applied to Indian languages. Direct translations often sound awkward, incomplete, or even misleading.
The risk is not just technical failure, but social exclusion.
“If AI becomes the main interface for services, and it only works well in a few languages, then we are creating a new kind of digital inequality,” Bhattacharyya warns.
The Private Sector Steps In
Companies like Reverie Language Technologies are trying to close that gap from the private sector side. Reverie’s AI-driven translation tools are now used by e-commerce platforms, fintech companies, logistics firms, and customer support systems across India.
But even as progress accelerates, there are concerns.
“AI can make communication easier, but it can also unintentionally marginalize less common dialects,” says Reverie co-founder Vivekananda Pani. “If we’re not careful, the languages with more data and commercial demand will dominate, while smaller dialects fade from digital use.”
That tension—between scale and inclusivity—lies at the heart of India’s AI language challenge.
Bhashini: The State’s Big Bet on Language AI
Recognizing the scale of the problem, the Indian government launched Bhashini in 2022 as part of the Digital India initiative. The goal is ambitious: to build the datasets, AI models, and translation tools needed to support all 22 official Indian languages.
Unlike global platforms that adapt models built elsewhere, Bhashini is explicitly India-first.
“Bhashini ensures India’s linguistic and cultural representation by building India-specific AI models,” says Amitabh Nag, CEO of Digital India’s Bhashini Division. “We are not relying on global platforms to do this for us.”
The numbers are striking. Bhashini currently hosts around 350 AI-based language models, which together have processed more than one billion tasks. Over 50 central government departments and 25 state governments are already using its tools.
Applications range from multilingual chatbots for public services to automatic translation of government schemes into local languages. The aim is to ensure that citizens can interact with the state in a language they understand.
Nag envisions a near future where rural users can access government services, financial tools, and information systems through voice-enabled interfaces in their native languages.
“If someone can speak to the system instead of filling forms in English, that changes everything,” he says.
AI Beyond Translation: The Next Frontier
While translation is a critical first step, researchers argue that the real power of multilingual AI lies in its ability to understand context, emotion, and intent.
At IIT Mumbai’s Koita Centre for Digital Health, Associate Professor Kshitij Jadhav is working on an AI system designed to help people quit smoking. The challenge, he says, is not just linguistic, but psychological.
“People at different stages of quitting need different kinds of conversations,” Jadhav explains. “You need empathy, assessment, and personalization. Traditionally, that requires a trained human.”
India simply does not have enough practitioners—especially multilingual ones—to meet the demand. Jadhav hopes AI can help bridge that gap.
The system he is developing will analyze how a user speaks, identify what kind of support they need, and tailor its responses accordingly—showing empathy, asking the right questions, and offering guidance.
“And all of this,” he says, “eventually needs to happen in 22 languages.”
Initial trials are underway in English and Hindi, but the long-term vision is far broader.
“It won’t be off-the-shelf AI,” Jadhav emphasizes. “It will be deeply customized for Indian users.”
The Cultural Stakes of Multilingual AI
Beyond efficiency and access, there is a deeper cultural dimension to India’s language AI push.
Language is not just a tool for communication; it carries identity, emotion, and worldview. When digital systems privilege certain languages, they implicitly privilege certain cultures.
That is why experts stress that multilingual AI must go beyond literal translation.
“Understanding idioms, social norms, and emotional cues is critical,” Professor Bhattacharyya says. “Otherwise, the technology may technically work, but it will feel alien.”
Projects like Bhashini aim to preserve that cultural richness by ensuring that AI models are trained on Indian contexts—not just translated versions of global data.
From the Lab to the Street
For people like Vineet Sawant, these debates play out in everyday life.
Since the translation feature was introduced, his work has become smoother, faster, and less stressful. He no longer hesitates when reading customer instructions or navigating app prompts.
“It gives confidence,” he says. “When the app speaks our language, we feel respected. We work better.”
His story is echoed by millions of workers, students, farmers, and small business owners across India who are encountering AI not as an abstract technology, but as a daily companion.
The Road Ahead
Making AI truly multilingual is not a one-time achievement—it is an ongoing process that requires constant data collection, refinement, and cultural sensitivity. The risk of uneven progress remains real, as does the danger that market forces could sideline less profitable languages.
Yet India’s approach—combining government-backed infrastructure like Bhashini with private-sector innovation—offers a model that other multilingual societies are watching closely.
If successful, India could demonstrate that AI does not have to homogenize human experience. Instead, it can amplify linguistic diversity—making technology more inclusive, more humane, and more effective.
As Sawant puts it, in simple but powerful terms: “Not everyone understands English. When technology understands us, we feel like we belong.”
In that sense, teaching AI to work in 22 languages is not just a technical challenge. It is a statement about who gets to participate in the future—and in what language that future will speak.