Indian Large Language Model Launched by Krutrim, the AI Startup of Ola Founder Bhavish Aggarwal
India’s inaugural multilingual large language model, capable of producing text in 10 Indian languages, has been introduced by Krutrim, an artificial intelligence startup established by serial entrepreneur Bhavish Aggarwal.
“LLM is voice-enabled and can understand multiple languages and even mixed languages like Hinglish-Hindi and English,” Aggarwal said at an event in Bangalore on Friday at the campus of Ola Group, where he is the head. managing director. “It is uniquely Indian.”
Krutrim, which means “artificial” in Sanskrit, also develops data centers and eventually aims to create servers and supercomputers for an artificial intelligence ecosystem. The chatbot will be available as a beta version in January. The prototypes of the servers will be completed in mid-2024 and production will begin at the end of 2025, the startup said in a statement.
A clutch of Indian startups and academic groups are racing to build large Indian language models, so-called Indian LLMs, since OpenAI’s ChatGPT launched a year ago. Countries hope to build their own competing AI systems instead of relying on US or Chinese technology. In Europe, investors are adding money to France’s Mistral AI, which is now valued at $2 billion after it was founded earlier this year. The United Arab Emirates promotes its Falcon model, which is supported by the Abu Dhabi Government Research Institute.
India, home to 1.4 billion people, is focusing on building smaller, more cost-effective AI systems. Generative AI startup Sarvam, which built its system using available open source models, launched OpenHath, the first open source Hindi LLM, earlier this week. The announcement comes days after it raised a $41 million investment from Lightspeed Venture Partners, billionaire Vinod Khosla and others.
At the event, Aggarwal was given the open source Krutrim template to welcome guests in English, write a poem in Tamil, compose an ode to monsoons in Bengali, and generate software code. “The AI models known around the world are largely trained in English,” he said. “They cannot capture our culture, our language and our ethos.”
He said the company is also focusing on chip development, including a “multi-chip” strategy, which it said will reduce costs and streamline data center design.
Krutrim – which is widely used by Ola Group’s ride-hailing company for voice chat, sales calls and customer support messages – also plans to introduce a business model called Krutrim Pro in the next quarter. Aggarwal said he uses the software to write performance reviews for his team and create job descriptions for hiring.
