India’s AI scene is on an upswing, but investors are cautious about which company to place their bets on, with most of the funding heading towards consumer tech and AI platforms built on existing foundational models like LLaMA and Mistral. Interestingly, there are not many research-based startups in India that’s focused on creating the engine on which AI apps are built. While India may seem to have arrived late to the AI party, it’s never too late to begin, with a few startups leading the way, taking on tech giants, and creating foundational models that are built on India-specific data sets.
KOGO OS
KOGO, Bengaluru-based startup, recently launched a human-first AI operating system built on large agentic frameworks, is designed to help developers build, deploy and manage autonomous AI Assistants that perform real world tasks. KOGO’s Large Agentic Framework is a new human-computer interface model that understands human intent from natural language.
KOGO OS can “understand”, “see” and “do” in twelve languages. With over 600 connectors to popular services like databases, API’s & services, helping developers build any kind of AI agent. KOGO can talk to any data and perform any task across multiple apps and platforms with accuracy. With a robust data governance layer, KOGO keeps data secure. The company’s proprietary AgentGraph, amplifies and augments businesses at significantly lower costs. KOGO OS is offering a marketplace for developers to build on top of its operating system, resulting in easy plug and play AI applications.
Sarvam AI
Sarvam AI, another Bengaluru-based startup, launched an open-source foundational model supporting 10 Indian languages, including Hindi, Tamil, and Bengali earlier this year. The company also introduced generative AI services for enterprises and released Sarvam 2B, an open-source 2-billion-parameter Indic large language model, trained on a proprietary dataset of 4 trillion tokens.
Sarvam AI’s platform is designed to be voice-enabled and multilingual, and it is a step in the right direction towards making AI accessible to people across India. The platform’s capabilities address the unique challenges posed by India’s linguistic diversity, ensuring that AI technologies are inclusive.
Krutrim AI
Krutrim is built on India-specific data. Starting with this AI model, the company plans to build other AI models across text, voice, and vision, and even go beyond large language models (LLMs) over time. Krutrim Pro, a large multimodal model, is under R&D, and is expected to have more sophisticated problem-solving and task-execution capabilities. Krutrim is currently ride sharing app Ola’s foundational model. Krutrim took three months to train its model’s first version, which is built on 2 million Indic tokens. This foundational model is by far the largest representation of Indian data used in training AI models and has an Indian persona.