Sarvam AI launches platform to help build India-centric AI models
Bengaluru, July 30
Bengaluru-based startup Sarvam AI on Thursday launched Epoch Builder Edition, a developer and enterprise platform aimed at helping organisations build, fine-tune and deploy large language models designed for Indian languages and use cases.
The company said the platform provides developers, researchers and enterprises with access to base models, training infrastructure, datasets and deployment tools, with a focus on reducing dependence on overseas AI models for Indian applications.
According to the company, the platform features new 7-billion and 70-billion parameter multilingual large language models (LLMs), trained on 2 trillion tokens and supporting more than 10 Indian languages, including Hindi, Kannada, Tamil, Marathi and Bengali.
The platform also offers GPU clusters for distributed training, reinforcement learning from human feedback (RLHF), instruction tuning, API access and on-premises deployment, the company said.
Sarvam AI said the platform provides curated datasets covering areas such as Indian law, governance, finance and vernacular content. It also includes tools for red-teaming, toxicity checks and benchmarking designed for India's cultural and linguistic context.
The company said the platform is aimed at banks, government departments, startups and research institutions looking to develop AI applications using their own data.
Speaking at the launch event, Sarvam AI Co-founder Pratyush Kumar said, "It is important that we have our own ability to build models that we need for a variety of applications... India has succeeded in building compelling global companies that have built in India and taken for the world, and the same can be absolutely done for AI."
The company said the platform aims to improve the performance of AI models on code-mixed languages such as Hinglish and Kannanglish, while offering on-premises and hybrid deployment options to help enterprises lower costs and retain control over their data. It also announced partnerships with three IITs and two state governments for pilot projects in education and public service delivery.
Sarvam AI said Epoch Builder Edition will be available in private preview from August 2026, with general availability planned for the fourth quarter of 2026.
— ANI
Reader Comments
This is impressive but I'm a bit skeptical about the timeline - private preview from August 2026? That's two years away! By then, OpenAI and Google would have released even more advanced models. The idea of India-centric AI is commendable but execution speed matters too. Let's see how it goes.
As someone working in a small govt department, this is a game-changer. The mention of curated datasets for Indian law, governance, and vernacular content is exactly what we need. We've struggled with generic AI tools that mess up local terminology. On-premises deployment option is a big plus for security-conscious organisations like ours.
Great to see this initiative! But I wonder if they've addressed the unique challenges of code-mixed languages like Hinglish? And what about regional dialects within each language? Also, hope they include more Indian languages beyond just 10. We have 22 scheduled languages after all. Still, it's a solid start. 👏
As an expat working in Bengaluru, I'm excited about this. The AI ecosystem in India is buzzing and this platform could help a lot of local startups. The partnerships with IITs are a smart move. One question though - will the base models be open-source or proprietary? That'll determine how much adoption they get globally.
Waah! This is what Atmanirbhar Bharat in tech looks like! 🇮🇳 Developing models that understand our cultural context, code-mixed languages like Hinglish, and local datasets for law and governance - this addresses real Indian needs that global models just can't handle. Kudos to the
We welcome thoughtful discussions from our readers. Please keep comments respectful and on-topic.