How to get AI to work in 22 languages

Vineet Sawant has spent the past two years navigating Mumbai’s busy streets on his scooter, delivering groceries to customers across the city.

“Being on the road is always very stressful, especially in cities like Mumbai,” he says.

But for Sawant, the challenge wasn’t just traffic — it was also language.

His first language is Marathi, and he speaks “very little” English. “I can understand, but reading it is very difficult,” he explains.

This created problems in his new job.

“At first, it was tough. Everything was in English. I could understand some of it, but I’m much more comfortable in Marathi. I used to ask other delivery guys to help me figure out what to do,” he recalls.

Working for Zepto — a company that promotes itself as offering “India’s Fastest Online Grocery Delivery” — meant clear communication was crucial. But drivers struggling to interpret English delivery instructions slowed things down.

To address the issue, Zepto partnered with Reverie Language Technologies about a year ago to introduce an AI-powered translation service for its drivers.

Since then, drivers have been able to choose between six languages directly on the Zepto app.

“I don’t have to guess anymore,” says Sawant.

“Earlier I would take more time to read and sometimes even made mistakes. Now, if the customer writes ‘ring bell’, I get that instruction in Marathi. I don’t have to ask or check again. It’s all clear.”

India’s Linguistic Diversity and the AI Challenge

India is home to 22 official languages and hundreds of dialects.

“Without technology that understands and communicates in these languages, millions are excluded from the digital revolution — especially in education, governance, healthcare, and banking,” explains Professor Pushpak Bhattacharyya from IIT Mumbai, a leading expert in AI for Indian languages.

The rollout of advanced generative AI systems like ChatGPT has made this work even more urgent. These systems require huge datasets — such as websites, books, and video transcripts — for training. While such data is readily available for widely spoken languages like Hindi and English, it is scarce for many others.

“The main challenge in creating Indian language models is the availability of refined data. Coarse quality data is available, but it’s not of very high quality — it needs filtering,” says Professor Bhattacharyya.

“In India, for many languages — especially tribal and regional dialects — this data simply doesn’t exist or isn’t digitised.”

Balancing Technology and Language Preservation

Reverie Language Technologies is now rolling out its AI-powered translation services for a variety of Indian businesses.

Co-founder Vivekananda Pani says that while AI translation will make communication smoother, there is a risk that lesser-known dialects could be left behind.

“The challenge will be to ensure that the incredible benefits of AI-driven language advancements don’t accidentally shrink the rich variety of human language,” Pani notes.

For drivers like Sawant, however, the technology has been life-changing — turning confusion into clarity, and helping them serve customers faster and more efficiently.

🌐 Read more news and updates here:
Digital More | Pakistan Rang