Bridging Language Gaps: AI Transforms Delivery for Mumbai Drivers

Sun Oct 19 2025 00:05:05 GMT+0300 (Eastern European Summer Time)

AI translation technology enhances communication for delivery drivers in multilingual India.

In a bid to overcome language barriers, Zepto partnered with Reverie Language Technologies to implement AI-powered translation services, enabling delivery drivers to receive instructions in their native languages. This initiative aims to facilitate seamless communication in a linguistically diverse India, where 22 official languages coexist. Experts emphasize the importance of high-quality language datasets for developing AI models, while ongoing projects like Bhashini strive to ensure that rural populations gain easy access to essential services in their mother tongues.

Vineet Sawant, a delivery driver in Mumbai, faced significant language challenges while navigating the city's bustling streets. With Marathi as his first language and limited proficiency in English, analyzing delivery instructions proved to be a daunting task. Recognizing the need for better communication tools, Zepto partnered with Reverie Language Technologies last year, launching an AI-powered translation service. This innovative solution allows drivers like Mr. Sawant to access instructions in their preferred languages directly through the Zepto app.

"Earlier, I would take more time to read and sometimes even made mistakes," admits Mr. Sawant. Thanks to the new feature, he can receive customer instructions in Marathi, enhancing his efficiency and reducing errors. Professor Pushpak Bhattacharyya from IIT Mumbai highlights the broader implications of such technology in India, where a spectrum of 22 official languages and numerous dialects exist.

The advancement of generative AI systems has underscored the urgency for language models that cater to these diverse dialects, which often struggle with data availability. Professor Bhattacharyya explains that while data exists for major languages, refining and digitizing data for many regional and tribal dialects remains a significant barrier.

In response to this challenge, Reverie Language Technologies aims to roll out its AI translation technology across various Indian firms. However, the co-founder, Vivekananda Pani, warns against the risk of sidelining lesser-known dialects as translation technologies proliferate.

To address this issue, Professor Bhattacharyya is actively involved in Bhashini, a government initiative focused on building high-quality datasets required for training AI systems. Launched in 2022, Bhashini has made remarkable progress, developing 350 AI language models that have handled over a billion tasks across multiple departments and state governments.

Amitabh Nag, CEO of Digital India’s Bhashini Division, emphasizes the necessity for unique India-centered AI models to ensure linguistic representation. He predicts that within the next couple of years, rural users will have seamless access to government services in their native languages.

Moreover, advanced AI models are being developed to tackle other pertinent issues such as smoking cessation. Kshitij Jadhav, associated with IIT Mumbai, is creating an AI program designed to provide personalized support in quitting smoking, with capabilities tailored to users' needs in multiple languages.

Ultimately, enhancements like these will allow drivers such as Mr. Sawant to deliver more efficiently, from an initial ten parcels to around thirty a day. "It makes us feel like we belong," he shares, reflecting on the newfound confidence that comes from working in one's preferred language. "When the app speaks our language, we feel more confident, and we work better."