How Father Solves Son’s Kannada Homework is a free keyboard app for 21 Indian languages

Bermi did what most people in his position would do: He opened the Google input tool in a browser. The web tool, which hasn’t changed much since its launch, lets you type phonetically in English and see Kannada characters appear on the screen. “But whenever I use it on my phone, it’s still not mobile-friendly,” he says. It required an internet connection and felt, in his words, “still stuck in that era”.

So he created an app. He took the Google Input Tools API, put it into a native iPhone app, and used it for about a year. Eventually, he decided to put it on the App Store. That’s when a small but consequential detail emerged. When the developers asked the developers what data their apps were collecting, the proprietary nutrition label realized that although Permi doesn’t store anything personally, every transcription query goes to an external server.

“It didn’t feel right, because I had no control over what data was being sent to Google. The keyboard also needed an Internet connection to work. At the same time, I started looking for an alternative on a completely personal, offline, device,” he says.

Keyboards and privacy

When it comes to the software layer, keyboards occupy an unusually sensitive position, unlike most applications where they sit between the user and everything else, logging every letter typed. Third-party keyboard developers have access to an optional ‘full access’ mode, which enables additional features but, in theory, data transfer. It is difficult to gain user trust and easy to lose.

“People use keyboards for many personal things like OTPs, credit card numbers,” he says. “I just wanted to assure you that I didn’t collect anything.”

Permi’s search for a privacy-preserving alternative led him to AI4Bharat, a research initiative based at IIT Madras that focuses on natural language processing for Indian languages. Among their open source releases was IndicXlit, a transliteration model trained in 21 Indian languages ​​that converts English input into native scripts.

Bermi shared that while IndicXlit is used in enterprise environments, he couldn’t find any consumer-level applications that integrated it. That’s when he decided to build one. “I felt a little happy and emotional because when you keep hearing about models being used at the consumer level somewhere outside India, when you see something that does everything on the device and integrates on the keyboard, without sending any information to any server or API, it feels like a breakthrough. It was great,” he shared.

Engineering solution

When asked how the AI4Bharat IndicXLit model works entirely on the device, Permi admitted that there were some challenges along the way. Keyboard extensions in iOS have strict memory limits, around 50 to 75 MB. The IndicXlit model, released, came in at approximately 350 MB and contained approximately 15 million parameters.

To fit it inside the keyboard, the Permi had to shrink significantly without reducing its output quality. This technique, called scaling, essentially reduces the numerical precision of a model’s weight and therefore requires less memory while preserving most of their behavior. Less accurate, but still meaningful, and much lighter.

“You can shorten it without losing any precision,” he said. “That’s when I put it on the keyboard and made sure the size was actually usable.”

Akshar allows users to type phonetically in English and instantly convert the text into Indian languages, all running entirely on the device. (Express Image)

The result is Akshar, a fully on-device keyboard that doesn’t require an internet connection and doesn’t collect user data. It supports 21 Indian languages, including many that don’t appear in competing tools like Konkani, Bodo, Kashmiri, Dogri, Maithili, Manipuri, Sanskrit and Sindhi. In comparison, Gboard supports six Indian languages ​​for transliteration; Apple’s own computer keyboard supports 11.

For speakers of languages ​​like Kashmiri or Bodo, the absence of prominent keyboards is no accident; This reflects a broader pattern in technology products, where support tends to follow user base size and business viability. “When any large company wants to add a feature, the question is, ‘Is there enough user base?’ Does the traction justify the effort?” says Permi. “As an independent developer, I can ship it, keep improving it, and put it in people’s hands.”

Before the interview, we tested the keyboard. Although it offers easy segue into scripts in different languages, a few nicks persist. When it comes to accuracy, Bermy admits it’s still a work in progress. He shared that the model generates multiple recommendations for each entry and spent considerable time building a custom dictionary layer to push the most common words to the top spot.

Spaces and journey to precision

Permi revealed that he tests each language with 20 to 30 real-world snippets, including mixed-language phrases that people actually type. Languages ​​like Kashmiri and Manipuri, which share alphabets with neighboring scripts, need special attention. For them, he keeps updating the dictionary based on user reports and his own testing.

There are gaps. Numbers are not yet cleanly integrated with transliterated text; Typing something like ‘drishyam 3’ in Hindi will give unexpected results. Autocorrect in Aksher is currently far from Apple’s standards, and Permi says that’s the next feature he’s working on: making the keyboard reliable enough in English that users don’t feel the need to switch to a computer keyboard.

Built using AI4Bharat’s IndicXlit model, Akshar supports 21 Indian languages ​​and works offline without sending user data to external servers. (Express Image)

A few weeks before this interview, Akshar launched in the App Store. When asked about the reception, Permi revealed that the feedback came from quarters he didn’t expect. A user in Maharashtra said he uses it to send WhatsApp messages to his business clients in Marathi. And when Permi posted about the app on Reddit, responses came from second-generation Indians in America—those who grew up speaking Indian languages ​​at home, could read and write them, and who spent their entire lives typing on English keyboards. “They said this is what they were looking for,” he recalled.

Free language app for everyone

The app is free, and Bermy says it will stay free. The economics are simpler than they seem: because everything runs internally via Core ML, there are no server costs, no API fees, and no scaling infrastructure. “I don’t have any hosting costs or server costs,” he says. “Core ML takes care of everything, runs everything internally.”

When asked if there are plans to bring the keyboard to Android, Permi said that it is under consideration. He works with Swift, an intuitive open-source programming language created by Apple, to build apps across iOS, MacOS, watchOS and tvOS. Official support for building Android apps using Swift was announced in October 2025.

Permi shared that he wants to bring Akshar to Android users — not least because the majority of Indian smartphone users are on that platform. He also keeps a close eye on AI4Bharat’s research outputs for new language models. Gondi, an officially recognized Indian language, is not yet supported by IndicXlit. He said that when a model is available, he will integrate it.

When asked what level of change he would like to see, the keyboard is a small step toward a broader vision. He sees technology as a way to help people communicate, learn and collaborate without being limited by the language they speak.