Sarvam partners EkStep, AI4Bharat to deploy multilingual voice AI agents across India

AI startup Sarvam AI has partnered with EkStep Foundation and AI4Bharat to deploy multilingual voice artificial intelligence (AI) agents across India under an initiative called “Listen at Scale.”

The project aims to shift public outreach from traditional one-way channels such as IVR calls and SMS to structured, two-way voice conversations in local languages.

By deploying conversational agents that operate in regional dialects, the initiative seeks to make digital services more accessible to populations that are typically excluded from text-based interfaces.


Over a 31-day period, the programme engaged around 50 lakh unique users across sectors such as healthcare, agriculture and governance, a report by the Bengaluru-based startup learnt.

The voice AI agents were used to automate various tasks, including beneficiary verification, citizen feedback and grievance redressal.

As part of the collaboration, EkStep Foundation provided the platform infrastructure and community outreach support.

Twenty selected teams received grants collectively worth Rs 1 crore, along with an allocation of five lakh voice AI minutes per team to build and deploy applications.

AI4Bharat, on the other hand, studied the results of the programme.

Sarvam AI use cases

The programme has already been used across several public sector initiatives. For instance, the National Health Authority used the system to reach over 14 lakh senior citizens, resulting in a 42% increase in daily enrolments for the Ayushman Vay Vandana Yojana.

The initiative also profiled over 4.14 lakh disabled persons for the Department of Empowerment of Persons with Disabilities, generating nearly 51,000 actionable profiles for assistive devices and financial aid.

Work with the Uttar Pradesh Skill Development Mission revealed that 92.6% of profiled informal workers had no prior formal work records, helping surface an otherwise “invisible” workforce.

State-wise impact

In Maharashtra, the AI system detected inefficiencies or possible middlemen interference in 2.75% of scheme delivery interactions.

Meanwhile, in Odisha, the programme confirmed a 77% seed receipt rate among farmers but identified an adoption gap in input procurement at 63%, providing insights for policy correction.

Sarvam recently open-sourced two reasoning models—Sarvam 30B and Sarvam 105B to support broader AI development.

Backed by investors such as Lightspeed Venture Partners, Peak XV Partners and Khosla Ventures, the startup is also supported by the government’s IndiaAI Mission and is preparing to raise additional capital to scale up its infrastructure.