Google Announces New AI Models, Tools to Boost AI Development in India

Google has announced a raft of new software and initiatives to accelerate AI-specific app and service development in India, including new models, a benchmarking tool, a hackathon, an association with the MeitY Startup Hub, and a programme granting up to $350,000 (approximately Rs. 2.92 crore) in Google Cloud credits. The announcements were made at the Google I/O Connect event in Bengaluru, a small offshoot of the company’s global-scale I/O developer conference. 

Starting with AI models and tools, Google announced the launch of IndicGenBench, a performance benchmarking tool designed specifically for local-language generative AI models. It covers 29 Indic languages in 13 scripts including many that have never been benchmarked before. This will allow developers to fine-tune language operations such as translating text and answering questions in different languages.

Project Vaani is a collaboration between Google and the Indian Institute of Science (IISc) to capture speech data for Indian languages. The company announced the completion of Phase I, which collected 14,000 hours of speech across 58 languages, spoken by 80,000 different individuals in 80 districts.   

Developers will also soon be able to easily build local-language versions of their generative AI apps using Google’s CALM (Composition of Language Models) framework, which the company has now open-sourced. 

All developers in India now have access to Google’s Gemini Pro 1.5 model with a context window of 2 million tokens, which generally works out to being able to process 2 hours of video, 22 hours of audio or massive amounts of text, allowing for extremely rich context in AI-generated material. Google’s smaller open-source Gemma 2 model is also now widely available, and the company is pitching it as particularly useful for multilingual apps thanks to its tokeniser which can break chunks of text down into smaller units for easier processing.

Other tools specifically tailored for India include Matformer, developed by Google’s DeepMind team in India, which will allow developers to mix and match different Gemini models to adapt to environments with varying resource availability on users’ phones. A new Agricultural Landscape Understanding Research API is aimed at those developing technology for the farming sector by taking advantage of remote sensing and hyperlocal mapping. 

The company has also cut pricing for Google Maps by up to 70 percent for Indian businesses, or up to 90 percent if they are participating in India’s Open Network for Digital Commerce. Google is also in the process of rolling out new features for Google Wallet so that developers can integrate loyalty programmes, tickets and gift cards. 

In collaboration with MeitY (the Ministry of Electronics and Information Technology), Google says it will be supporting 10,000 Indian startups that are working with AI. This will come in the form of up to $350,000 worth of Google Cloud credits, which will give developers access to cloud computing power and infrastructure, in addition to skill development and providing mentorship. A nationwide AI hackathon and an AI startup bootcamp are also being planned. 

In addition, three startups supported by Google showcased their work based on generative AI at the event. Miko.AI is developing a robot for children’s education using its own proprietary AI models. Karya is a platform to help those in low-income communities find work. Cropin is an agricultural intelligence platform that helps farmers optimise their work based on climate and food supply data.