India has 1.4+ billion people. Most government services, welfare schemes, and livelihood programmes are built for those who can read, type and navigate in English or Hindi. This excludes the majority of people that these services are designed to reach.

The barrier is not education or intent. It is the interface.

Voice AI removes that obstacle entirely. A phone call in any language is sufficient. The technology works. The cost has come down to <₹1 per minute for a full end-to-end stack.

The question for any organisation working in public services is no longer whether Voice AI works. It is how quickly it can be adopted - and what it takes to deploy it for a specific use case, in specific languages, at a cost that public systems can sustain.

This ecosystem exists to answer that question.

25+

citizen-facing deployments

1000+

1000+ community members

<₹1/min

full stack cost benchmark

Video

What EkStep Does in Voice AI

The most common reason Voice AI projects stall is not technology. The technology works. The models are good enough. The cost is manageable.

What stalls adoption is a lack of clarity on what has already been built, deployed where it failed, what it costs from end to end and how to restart without rebuilding from scratch.

EkStep Foundation’s Voice AI work addresses this gap. Across agriculture, livelihoods, health, and governance, it has seeded open infrastructure, run population-scale deployments, convened the practitioner community, and published learning documents so that organisations can learn from each other.

Use Case Deployments

Where Voice AI is working - in production, at scale

Agriculture

India’s agricultural advisory system has always had a last-mile problem. Extension workers cannot reach every village at every critical moment in the farming calendar. Printed guides assume literacy. Call centres operate primarily in Hindi. The farmer who speaks Marathi, Odia, or Gujarati - and needs pest management advice on a Thursday afternoon or wants to know the current MSP for maize before deciding whether to sell - has had no reliable, accessible option.

Large-scale Voice AI deployments have changed that.

MahaVISTAAR

MahaVISTAAR was launched in May 2025 by the Government of Maharashtra in collaboration with EkStep Foundation as part of the VISTAAR Digital Public Infrastructure initiative. It gives farmers in Maharashtra real-time, personalised agricultural advice through voice calls in their own language - covering crop advisories, pest management, input guidance, and scheme information.

Sarlaben by AMUL

Sarlaben by AMUL, launched in February 2026, is an AI-powered advisory assistant developed in partnership with the EkStep Foundation. It serves 3.6 million dairy farmers across 18,600 villages in Gujarat. Farmers call and speak in Gujarati. They receive personalised advice on cattle health, breeding, and fodder. The app is based on five decades of Amul data and is delivered through voice calls and the Amul Farmer Mobile App. The farmers who are mostly women, do not require a smartphone, English literacy or technical knowledge.

Livelihoods

India’s local job markets do not have a supply problem. Jobs exist. Job seekers exist. They cannot find each other because neither is on a shared digital system. The employer puts a notice on a pole. The job seeker asks around. Both wait.

BlueDots, an EkStep Foundation initiative, uses Voice AI to make local labour markets discoverable. A job seeker calls, speaks for two minutes in any language, and becomes a precise, location-anchored signal on a shared district map. An employer describes a vacancy the same way. The system matches them.

In Ghaziabad, this approach surfaced 10,000+ local job openings from less than 10% of the district’s businesses ,while national platforms were showing fewer than 100 listings in the same area. For employers the time taken to source a relevant candidate dropped from over two weeks to under two days.

The Open Infrastructure

VoicERA

Built for Bharat. Scales to the World.

An open-source, end-to-end Voice Al infrastructure for population-scale deployments. VoicERA brings telephony, speech models, LLMs, and agent orchestration together-delivering deterministic latency, high reliability and full control over data and performance.

Core Capabilities
Fully Open-Source Voice Stack

End-to-end, auditable, and extensible-no vendor lock-in

Interoperable by Design

Works across multiple telephony providers, ASR/TTS engines, and LLMs

Optimized for On-Prem & Cloud

Built for government and enterprise environments with strict data sovereignty needs

Low-Latency, High-Performance

Real-time streaming audio with predictable response times at scale

Hosted on National Infrastructure

Deployed on BHASHINI-National Infrastructure

Platform-Ready

In-built agent builder, prompt orchestration, analytics, and monitoring-from pilot to production

VoicERA Co-Creators

AI4Bharat

Voice AI for India’s 22 Languages

Open-source AI models powering voice assistants, real-time translation, and speech recognition across all 22 scheduled Indian languages - making technology accessible to India’s 1.4 billion people.

Speech-to-Text IndicConformer
IndicTrans3
Translation
Text-to-Speech IndicParler

450k+

Model Downloads

22

Indic Languages Supported

17k+

Speech Data Hours

MIT

Open-Source License

Healthcare

Education

Citizen Support

Applications & Services

Government Systems

The Partner Ecosystem

AI4Bharat

Open-source Indic language models, datasets, and tools. Research lab at IIT Madras.

Sarvam AI

Leading Indian AI company building foundation models for Indic languages.

Intel

Evals & Voice Harness — sandbox and benchmarking collaboration.

Anthropic

AI safety research. Collaboration on Trust, Safety & Moderation layer.

NVIDIA

GPU and edge optimisation for population-scale inference.

IIT Madras

Home of AI4Bharat. Indic language models, evaluation, benchmarking.

Ashoka University

Safety, cultural nuance, and evaluation for responsible AI in India.

Bhashini

India's national language translation mission. Hosts VoicERA. Core infrastructure partner.

Programmes

From passive broadcasting to population-scale conversation - what 31 days proved.

For decades, government outreach has meant IVR and SMS. Systems that push information to citizens without the capacity to receive anything back. Citizens hear. They cannot respond, verify, clarify, or signal that they need something different. The institution sends. It never knows if anything landed.

Listen At Scale

Hosted by the EkStep Foundation, with Sarvam as the technology partner and AI4Bharat as the knowledge partner, the ‘Listen At Scale’ programme tested a fundamentally different model. Conversational Voice AI agents spoke in local languages, asked structured questions, received real responses, and converted those responses into data that institutions could act on immediately.

Between January and February 2026, twenty organisations deployed Voice AI agents across healthcare, agriculture, governance, skilling, and civic engagement. Each organisation received a grant of 5 lakh Voice AI minutes to move from idea to live deployment.

The combined scale

7.4 million

Voice AI minutes

~5 million

Unique citizens reached

20

Organisations — state governments, central ministries, and non-profits

Impact At Scale

Hosted by the EkStep Foundation, with Raya, VocalLabs, and Samora.ai stepping forward as partners from the community, this programme tested a unique economic model. Conversational Voice AI agents spoke in local languages, handled real citizen interactions, and delivered measurable outcomes — at a cost that, for the first time, made scale genuinely within reach.

For a period of 100 days, any organisation could deploy Voice AI agents across any use case with no cap on minutes, and access was priced under ₹1 per minute, end-to-end.

The combined scale, live and growing

Unlimited

Voice AI minutes — no ceiling on deployment

50+

Active use cases across healthcare, agriculture, governance, skilling, and civic engagement

20+

organisations and counting

Meet the Voice AI Team

Santosh Kevlani

Head, Voice and Language AI

BalaKrishnan A

Senior Programme Manager

Pranava C Hiremath

Product Manager

Gallery

Events

Jan 28th, 2026

Voice AI: Making the Best Work for India

Practitioner gathering on deployment realities across India’s languages

Feb 15th-20th, 2026

India AI Impact Summit

Launch of VoicERA on BHASHINI, National Language Infrastructure

Ongoing

Voice AI Roundtables

Regular gatherings of deployers, builders, and researchers working through shared challenges (Voice AI Telephony Roundtable, Multilingual Support on National Helplines)

Resources and Documents

What the ecosystem has built and published - for every organisation deploying next
PDFs available for all five documents. Contact: voice@ekstep.org

How to Start Your First Voice AI

Guide · Beginner

Cloud Telephony Partner for Your How to Procure a Voice AI Deployment

Guide · Telephony

Understanding the Voice AI Tech Stack

Architecture · Intermediate

Open Voice AI Stack - Starter Kit

Tool · Open Stack

Voice AI Learnings from Real Projects

Learnings · All Levels

Community Stats

1000+

Voice & AI startups, founders, developers

50+

Social & non-profit enterprises on citizen AI

20+

VC partners in the ecosystem

100+

Researchers & academics from leading institutions

Join Us

Be part of building Voice AI for real-world use

Bring Voice AI to the people who need it most. Access programs, partners, and tools to move from idea to deployment.

Contact Voice@EkStep.org