Bengaluru-based Sarvam AI has launched Epoch Builder, a platform helping Indian organizations build custom AI models in over 10 local languages. The platform targets reduced reliance on foreign technology and includes tools for data curation and training. Private previews begin in August 2026, with full availability expected in the final quarter of the year.
Sarvam AI, a Bengaluru-based startup, has officially introduced its Epoch Builder platform, aiming to provide Indian enterprises and researchers with the infrastructure needed to create large language models (LLMs) focused on domestic requirements. The launch addresses a growing need for AI tools that understand the nuances of Indian languages and regional contexts, which are often overlooked by global AI models developed primarily in English.
The platform offers foundational models with 7 billion and 70 billion parameters, trained on two trillion tokens. These models are designed to support more than ten Indian languages, including Hindi, Bengali, Marathi, Tamil, and Kannada. A notable technical focus of the platform is its capability to handle code-mixed languages, such as Hinglish and Kannanglish, which are commonly used in digital communication across India.
Infrastructure and Deployment Features
To facilitate development, the Epoch Builder platform provides access to GPU clusters for training and tools for techniques like reinforcement learning from human feedback and instruction tuning. For enterprises concerned with data privacy, Sarvam AI has included options for on-premises deployment, allowing organizations to manage sensitive information internally rather than relying on external cloud hosting. Additionally, the platform provides curated datasets covering specialized fields like Indian finance, governance, and law, along with tools for benchmarking and safety testing.
Strategic Focus and Collaboration
Pratyush Kumar, Co-founder of Sarvam AI, noted that building indigenous AI capacity is essential for India to develop applications specific to its diverse population. The company has already initiated pilot collaborations with three Indian Institutes of Technology (IITs) and two state governments. These initial projects are centered on public service delivery and education, areas where localized AI could potentially improve efficiency and accessibility.
The company confirmed that the Epoch Builder Edition will move into a private preview phase in August 2026. A wider, general release is currently scheduled for the fourth quarter of 2026. For investors and industry observers, the key monitorable will be the adoption rate of this platform by large Indian corporations and government bodies, as well as the company’s ability to compete with well-funded global AI players who are also expanding their presence in the Indian market. While the platform offers advanced technical capabilities, the long-term success of such initiatives often depends on scalability, cost-effectiveness for local users, and the ongoing demand for specialized, language-specific AI solutions.