India: using AI to make government services available in 22 languages

The problem

India officially recognises 22 languages, but hundreds more are spoken across the country. Only 11% of the population speaks English, and even Hindi — the most widely spoken language — is used by just 57%. For the hundreds of millions of people who may not speak either language fluently, digital government services that are only available in English or Hindi are effectively out of reach — not because of a lack of phones or internet access, but because the technology is not available in their language.

Government websites, online forms, welfare applications, and legal information are typically available in English, Hindi, and sometimes one regional language. For anyone who does not speak those languages fluently, even a routine task — checking eligibility for a benefit, understanding a legal right, applying for a document — can be out of reach. The barrier sits between citizens and the services their government provides, and the people most affected are those with the fewest alternatives.

Building every government service in every language from scratch would be impractical for a country with India's linguistic diversity. What was needed was a shared foundation — a set of tools that any government agency, developer, or organisation could use to make their services work in multiple languages without each one building that capability independently.

The solution

In 2022, the Indian government launched Bhashini (short for “BHASHa INterface for India”). This national platform provides translation, speech recognition, and text-to-speech tools across all 22 constitutionally recognised Indian languages. Rather than translating individual government websites one by one, the government built the underlying language technology and made it freely available for anyone to use.

Bhashini is designed so that other organisations can plug into it. If a state government is building a welfare portal, or a nonprofit is creating a legal advice service, they can connect to Bhashini and immediately offer that service in multiple Indian languages — without having to develop their own translation or speech recognition technology. The language tools are shared, free, and ready to use.

Building those tools, however, required something that did not yet exist for many Indian languages: enough real-world language data for the AI to learn from. Translation and speech recognition systems improve by processing large volumes of examples — recordings of how people speak, written translations between languages, text in different scripts. For English and Hindi, this kind of data is widely available online. For many of India's other languages, it is not.

The government addressed this through BhashaDaan (“Language Donation”), a programme that invites ordinary citizens to contribute. Using a smartphone, anyone can take part: recording their voice reading sentences aloud, listening to audio and writing down what they hear, translating text from one language to another, or reading text from an image and typing it out. Each activity takes only a few minutes.

These contributions are collected through AIKosh, India's national repository of AI training data, and used to build ready-to-use language models hosted on a platform called Model Vatika, which developers can plug into their own applications. The result is a pipeline where citizens' language contributions become public infrastructure that benefits everyone.

The impact

1. Multilingual capability is available as shared public infrastructure across India Before Bhashini, any organisation that wanted to offer services in multiple Indian languages had to build or buy that capability independently. Bhashini provides translation, speech recognition, and text-to-speech across 22 languages through open interfaces that anyone can connect to for free. A small nonprofit, a state government department, or a village council can add multilingual support without the cost or complexity of developing language tools themselves.

2. Village councils are documenting proceedings in local languages
Thousands of village councils across India now use a platform called SabhaSaar, which runs on Bhashini's language tools, to record and manage meeting minutes. Officials speak in their local language, and the system automatically transcribes and translates the proceedings. Meetings that previously went undocumented or required manual translation are now captured and accessible in the language the community speaks.

3. Citizens are accessing legal rights and welfare information in their own language through services built on the platform
A WhatsApp-based assistant called Jugalbandi, built by the nonprofit OpenNyAI using Bhashini's tools, helps people navigate legal rights and government programmes through voice messages in their own language. A domestic violence survivor can send a voice message asking about their legal options and receive a response explaining their rights in the same language. Sanitation workers in Bangalore have used it to understand their labour rights. Farmers have used it to find out about agricultural subsidies they may be eligible for. Because Bhashini's tools are free and open, OpenNyAI was able to build this service without developing its own language technology.

4. State governments are beginning to integrate Bhashini into their own digital services
In January 2026, the state government of Madhya Pradesh signed a partnership with the Bhashini Division to integrate multilingual AI tools across its state digital governance platforms. This represents a shift from Bhashini being used by individual third-party applications to being embedded in how a state government delivers services to its citizens.

Lessons learned

• The government built shared infrastructure rather than translating services one by one. Bhashini is not a single government website made available in 22 languages. It is a set of language tools that any organisation can connect to and use. The government built the foundation and made it open; others build the services their communities need on top of it. For governments in other multilingual countries, this approach may be more practical and scalable than translating every service individually.
• The language data was built through public contributions. For many Indian languages, the data needed to train AI language tools did not exist in sufficient quantity online. BhashaDaan addressed this by inviting citizens to donate recordings and translations of their own languages, capturing how people actually speak rather than relying on formal written sources. For any country where AI language tools perform poorly in local languages, crowdsourcing from native speakers is a practical way to build the data those tools need.
• Making the tools free and open enabled applications that the government did not build itself. Jugalbandi was built by a nonprofit. The Sansad Bhashini parliamentary translation and the SabhaSaar village council platform serve entirely different purposes but use the same underlying infrastructure. When governments open up foundational technology, the range of what gets built on top of it can extend well beyond what the government would have designed on its own.

This case study was written with assistance from artificial intelligence.