Bengaluru, India – Sarvam, an Indian AI startup, is addressing a critical gap in the market with the launch of a series of AI tools tailored to the linguistic diversity of India. With the country's 22 official languages and over 19,000 dialects, Sarvam is betting on the increasing preference for voice interactions over text, particularly in customer support scenarios.
The startup unveiled its new voice-enabled AI bot, which supports more than 10 Indian languages. This move aligns with the belief that users are more inclined to engage in their native languages rather than struggle with typing in complex scripts. Co-founder Vivek Raghavan highlighted the challenge of typing in Indian languages and emphasized the preference for spoken communication.
Sarvam’s AI voice agents can be deployed across various platforms, including WhatsApp, mobile apps, and traditional voice calls. The company already boasts a successful implementation with Sri Mandir, a startup offering religious content, which has processed over 270,000 transactions using Sarvam's AI technology.
Sarvam is also introducing a small language model, Sarvam 2B, which is trained on a massive synthetic dataset of 4 trillion tokens. This model is designed to be cost-effective and efficient, with Sarvam claiming it will cost a fraction of comparable industry models. The startup is open-sourcing Sarvam 2B, encouraging community contributions to further enhance its capabilities.
Alongside Sarvam 2B, the company is launching Shuka, an audio-language model developed with Saaras v1 and Meta's Llama3-8B Instruct. This model, also open-sourced, offers translation, text-to-speech (TTS), and other modules to support the creation of voice interfaces.
Another significant addition is A1, a generative AI workbench for legal professionals. This tool can perform tasks such as looking up regulations, drafting and redacting documents, and data extraction.
Sarvam’s initiatives are in line with India's broader push towards sovereign AI development. The IndiaAI program aims to develop AI infrastructure that respects national data privacy and caters to local needs. Sarvam is poised to contribute to this initiative, particularly through the IndiaAI Compute Capacity project, which seeks to establish a supercomputer with 10,000 GPUs.
Raghavan expressed readiness to collaborate with the government if opportunities arise, highlighting Sarvam’s commitment to advancing India’s AI capabilities while addressing the unique linguistic and cultural needs of its diverse population.
