Sarvam AI has launched Epoch Builder Edition, a platform designed to help organisations build, fine-tune and deploy large language models for Indian languages and local use cases.
The Bengaluru-based company is positioning the platform as a common development environment for enterprises, startups, researchers and public institutions that want to create AI applications using their own data. Sarvam AI said users will be able to access base models, datasets, computing infrastructure and deployment tools through the platform.
At the centre of the offering are two multilingual language models with 7 billion and 70 billion parameters. According to the company, the models were trained on two trillion tokens and support more than 10 Indian languages, including Hindi, Kannada, Tamil, Marathi and Bengali.
Sarvam AI said the platform has also been developed to handle code-mixed language, where users switch between English and an Indian language in the same conversation. The company specifically cited Hinglish and Kannanglish as areas where it wants to improve model performance.
That focus could be important for organisations building services for Indian users, where everyday communication often differs from formal written language. Rather than limiting developers to a finished chatbot or a standard application interface, Epoch Builder Edition is intended to give them more control over how models are trained, tested and deployed.
The platform will provide access to GPU clusters for distributed training, along with reinforcement learning from human feedback, instruction tuning and application programming interfaces. Customers will also be able to choose on-premises or hybrid deployment, which the company said can help organisations retain control over their data and manage costs.
Indian datasets and local testing tools
Sarvam AI said Epoch Builder Edition will include curated datasets covering Indian law, governance, finance and vernacular content.
The platform will also offer red-teaming, toxicity testing and benchmarking tools designed around India’s linguistic and cultural context. These features are intended to help developers examine how a model behaves before it is used in public-facing or institution-specific applications.
Banks, government departments, startups and research institutions are among the groups the company expects to use the platform.
Organisations will be able to develop models and applications around their own information rather than relying entirely on externally developed systems.
Speaking at the launch, Sarvam AI co-founder Pratyush Kumar said India needed the ability to build models suited to a wide range of domestic applications.
“It is important that we have our own ability to build models that we need for a variety of applications,” Kumar said. “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 is part of an effort to reduce dependence on overseas AI models for Indian applications. Its approach brings together language models, computing resources, sector-specific datasets and deployment options in a single offering rather than treating them as separate services.
Pilot projects with IITs and state governments
Sarvam AI also announced partnerships with three Indian Institutes of Technology and two state governments for pilot projects in education and public-service delivery. The company did not identify the institutions or governments involved in the announcement.
The pilots are expected to provide early examples of how the platform can be used in academic and government settings, although further details about their scope were not disclosed.
Epoch Builder Edition is scheduled to enter private preview in August 2026. Sarvam AI plans to make it generally available in the fourth quarter of the year.
Also Read: iPhone 18 Leaks Point to A20 Chip, 12GB RAM
