Google is targeting India’s rapidly growing AI developer community by launching a suite of tools, programs, and partnerships to aid in creating AI-driven products for both domestic and global markets.
“India is pivotal to our global AI mission. With its large mobile-first population, vibrant startup ecosystem, and diverse linguistic landscape, we’re uniquely positioned to drive global AI innovation,” said Seshu Ajjarapu, Senior Director at Google DeepMind.
India is leading in the adoption of Google’s Gemini family of AI models, utilized by over 1.5 million developers worldwide. He added that India boasts one of the largest user bases on Google AI Studio, a platform for rapid prototyping and experimenting with generative AI models.
Manish Gupta, Director at Google DeepMind, highlighted three focus areas for the company: multimodal, multilingual, and mobile AI. “We are working globally to incorporate these capabilities into Gemini,” he said. On July 17, during a developer event in Bengaluru, Google announced the release of Gemma 2, the next generation of its open-source AI model, to developers in India.
Originally introduced at the Google I/O developer conference in the US in May, Gemma 2 features a new architecture for improved performance and efficiency, available in 9 billion and 27 billion parameter sizes. The model’s tokenizer is particularly effective for building multilingual solutions for India’s diverse languages.
This was demonstrated by Navarasa, a multilingual variant for Indian languages built on Gemma by Telugu LLM Labs, a collaboration between LlamaIndex’s Ravi Theja Desetty and Knownwell’s Ramsri Goutham Golla. Navarasa supports 15 Indian languages and aims to enhance the AI experience for Telugu speakers worldwide.
Google is also providing a 2 million token context window on its flagship Gemini 1.5 Pro model to all developers in India. This context window size determines how much data a model can process at once, enabling it to handle up to 1 hour of video, 11 hours of audio, or extensive codebases and text up to 1,500 pages.
To support high-quality language model development for India’s linguistic diversity, Google introduced IndicGenBench, a multilingual benchmark suite for Indian languages. Developed by Google DeepMind’s India unit, it evaluates language generation capabilities across 29 Indian languages.
Additionally, Google is open-sourcing CALM (Composition of Language Models), allowing developers to combine their specialized language models with Gemma models for more nuanced solutions. This technology enhances language inclusivity in AI models, enabling developers to mix capabilities from different models.
Gupta explained that the DeepMind India team developed a small model called Morni, proficient in Indian languages, which they combined with the powerful Gemini model. This combination aims to deliver the best of both worlds in AI performance and linguistic understanding.
Furthermore, the DeepMind India team contributed to developing the Matformer framework, enhancing on-device AI capabilities. The framework will be included in the upcoming Gemini Nano version, allowing developers to optimize for performance and resource consumption on users’ phones.
Project Vaani, a collaboration between Google, the Indian Institute of Science (IISc), and ARTPARK, completed its first phase, providing over 14,000 hours of speech data across 58 languages. The project aims to collect and transcribe speech data from all 773 districts of India in three phases, with phase two currently underway.
Google also plans to launch an Agricultural Landscape Understanding (ALU) Research API in limited preview to make agricultural practices more data-driven. This API aims to help farmers access subsidies, improve yields, and enhance market access by leveraging AI and remote sensing.
Additionally, Google has introduced India-specific pricing for developers using the Google Maps platform, offering up to 70 percent lower costs on most APIs. The company has also partnered with the Open Network for Digital Commerce (ONDC) to offer up to 90 percent off on select Google Maps Platform APIs for developers building for ONDC.
