DEV Community: Adarsh Kant

Why Indian Businesses Are Replacing Chatbots with Voice AI in 2026

Adarsh Kant — Fri, 27 Feb 2026 20:06:07 +0000

India has 800 million internet users. Most of them prefer speaking over typing.

Yet every chatbot on every Indian website assumes users will type fluently in English on a tiny phone screen. That's broken.

The Problem

Here's what happens when a Tier-2 city customer visits a D2C website:

They look for a phone number (not found)
They try the chatbot, type "return kaise karein" (Hindi-English mix)
Chatbot doesn't understand
They leave. Forever.

You just lost ₹2,000 because your chatbot couldn't understand spoken Hindi.

The Voice-First Fix

Voice AI on websites is fundamentally different from chatbots:

Feature	Chatbot	Voice AI
Input	Typing only	Voice + Text
Language	English-first	Hindi, Hinglish, 10+ languages
Navigation	Links & buttons	Speaks AND navigates your site
Memory	None	Remembers returning visitors
Mobile UX	Tiny keyboard	Just speak naturally

It's NOT a Phone Bot

Important distinction: voice AI platforms like Vapi and Bolna are phone call tools. They're great for IVR replacement and outbound calls.

But website voice AI is a different category. It lives on your site, talks to visitors, navigates pages, fills forms, and remembers users across sessions.

Who's Using This?

D2C brands — "Show me kurtas under ₹1,500" (voice-guided product discovery)
EdTech — Course counselling in Hindi for Tier-2/3 students
Healthcare — Appointment booking for patients who can't navigate complex UIs
Real estate — "3BHK near metro, under 80 lakhs"

Try It

I'm building AnveVoice to solve this exact problem. One line of JavaScript, 5-minute setup, free tier included.

The next 500 million Indian internet users won't type their way through your website. They'll speak.

Try the live demo →

Built in India, for Indian businesses first. Would love feedback from the DEV community!

Why Voice AI is Replacing Chatbots for Indian SMBs

Adarsh Kant — Mon, 23 Feb 2026 08:19:18 +0000

Why Indian SMBs Are Ditching Chatbots for Voice AI

The problem: 85% of Indian internet users are mobile-first. Typing on phones is painful, especially in regional languages.

What we're seeing:

3x higher engagement with voice vs chatbots
40% longer session times
Customers actually enjoy the interaction

Real use cases:

Dental clinics booking appointments via voice
D2C brands handling returns through voice
Edtech platforms guiding course selection
Real estate qualifying leads

The shift: Forward-thinking SMBs aren't upgrading chatbots. They're replacing them entirely.

Because when you remove the need to type, you remove the friction.

Try it: https://anvevoice.app/demo?utm_source=devto&utm_medium=article&utm_campaign=feb23

Built for Indian SMBs. 22 languages. 30-second setup.

voiceai #india #smb #startups

I Built an AI Voice Assistant for My Dad's Dental Clinic — Here's What Happened

Adarsh Kant — Sat, 21 Feb 2026 16:36:56 +0000

Three months ago, my dad's dental clinic was losing 40% of potential patients. Not because of bad service — because of missed calls.

Patients would call after 6 PM, during lunch breaks, or on Sundays. The clinic's voicemail would pick up, but most people don't leave voicemails anymore. They just call the next clinic.

The Problem

My dad runs a successful dental practice in Delhi. Great reviews, loyal patients, modern equipment. But his front desk could only handle calls during business hours (9 AM - 6 PM).

The math was brutal:

60% of calls came during business hours ✓
40% came after hours, weekends, holidays ✗
Most of those 40% never called back

Each missed call = ₹3,000-₹5,000 in lost revenue (average dental appointment value).

The Solution

I built AnveVoice — an AI receptionist that actually talks to website visitors.

Not a chatbot. Not a form. A real voice conversation.

What it does:

Answers questions in Hindi, English, Tamil, Telugu (22 Indian languages)
Books appointments directly via Calendly integration
Captures lead information 24/7
Fills out patient intake forms automatically

The Results (30 Days In)

40% more appointments booked — mostly after hours
60% reduction in "how do I..." phone calls
Zero missed calls after 6 PM
Patients actually prefer it — especially younger ones

Technical Details

Built with:

OpenAI GPT-4 for natural conversations
46 MCP tools for full platform control
Calendly API for appointment booking
WebRTC for browser-based voice

NEW: Just shipped OpenClaw skills + full API for developers.

Try It

If you're running a clinic, salon, consultancy, or any appointment-based business:

👉 anvevoice.app — Free tier available

Would love your feedback. What would you want an AI receptionist to handle?

Why Indian SMBs Are Switching to Voice AI (And Why You Should Too)

Adarsh Kant — Sat, 21 Feb 2026 00:14:47 +0000

Last week, I watched a dental clinic owner in Delhi struggle with the same problem every SMB faces: customers visit the website, start a chat, then disappear. The chatbot gets 2% engagement. The other 98%? Gone forever.

Here's what's changing in 2026: Indian SMBs are replacing chatbots with voice AI — and seeing 3x more engagement.

The Problem with Chatbots

Chatbots seemed like the answer. They're cheap, easy to set up, and available 24/7. But here's what actually happens:

2% engagement rate — Most visitors ignore them completely
Menu fatigue — Users get stuck in button trees, frustrated after 3 clicks
Mobile nightmare — Typing on phones while browsing? Nobody wants that
Language barriers — English-only chatbots alienate 90% of Indian users

I talked to 50+ SMB owners last month. The sentiment was unanimous: "Chatbots feel robotic. Our customers don't use them."

Why Voice AI Wins

Voice AI changes the equation completely:

15-20% engagement — People actually talk to it
Natural conversations — No menus, no buttons, just ask
Hindi + English — Code-switching is native, not an afterthought
Mobile-first — Perfect for India's 850M+ smartphone users
5-minute setup — One script tag, no complex flows

Real Results from Real Businesses

A D2C brand in Mumbai added voice AI to their Shopify store. Result: 35% more conversions, 60% fewer support tickets.

A dental clinic in Bangalore replaced their chatbot. Result: 40% of appointments now booked after hours, when no receptionist was available.

An edtech platform integrated voice AI for student queries. Result: 3x more engagement, students asking 5x more questions.

The India Angle

This matters more in India than anywhere else:

63M+ SMBs — Most don't have tech teams
Voice-first users — Next 500M internet users prefer speaking over typing
22 languages — Voice AI handles multilingual natively
Mobile-only — 80% of Indian internet is mobile

Making the Switch

The best part? Migration takes 5 minutes. Remove your chatbot script, add voice AI, done. No retraining. No flow building. It learns from your website automatically.

If you're still using a chatbot in 2026, you're leaving 90% of your visitors behind.

What's your experience with chatbots vs voice? Drop a comment below.

How We Built a Voice AI That Takes Real DOM Actions

Adarsh Kant — Fri, 20 Feb 2026 00:58:04 +0000

Voice is the new click. But most voice AI today? It's just chatbots with a microphone.

The Chatbot Trap

Traditional voice assistants follow a familiar pattern: listen → transcribe → think → respond. The problem? They're passive. They can tell you how to do something, but they can't actually do it.

When a user says "book me a flight to New York," they don't want a list of booking sites. They want a confirmation email.

Voice-First Architecture

We built Anve differently. Instead of just generating text responses, our AI makes decisions about actual DOM actions it can take on the user's behalf.

The flow looks like:

Intent Recognition - What does the user want?
Action Planning - What DOM operations achieve this?
Execution - Actually click, type, and submit
Confirmation - Verify the action succeeded

The DOM Action Engine

This is where it gets interesting. Our AI doesn't just see your website—it can interact with it.

// The AI generates action sequences like:
[
  { type: 'click', selector: '#book-flight' },
  { type: 'type', selector: '#destination', value: 'New York' },
  { type: 'click', selector: '#search' }
]

We use a custom-trained model that understands:

CSS selectors and element hierarchies
Form validation states
Async operation timing
Error recovery patterns

5-Minute Integration

Getting started is surprisingly simple:

React/Next.js:

import { AnveVoice } from '@anve/voice-react';

function App() {
  return (
    <AnveVoice 
      apiKey={process.env.ANVE_API_KEY}
      actions={['navigate', 'click', 'type', 'submit']}
    />
  );
}

Vanilla JS:

<script src="https://cdn.anvevoice.app/voice.min.js"></script>
<script>
  Anve.init({
    apiKey: 'your-api-key',
    container: '#voice-widget'
  });
</script>

Live Demo

See it in action: anvevoice.app/demo

The demo shows a real booking flow completed entirely through voice commands. No pre-scripted responses—just actual DOM manipulation happening in real-time.

Want to add voice to your app? We're opening early access for developers who want to build voice-first experiences. Join the waitlist

I Replaced My Website's Chatbot with Voice AI — Here's What Happened

Adarsh Kant — Thu, 19 Feb 2026 08:10:44 +0000

Last month I ripped out the chatbot widget from our website. My co-founder thought I'd lost it.

"We spent three months building that chatbot," he said. "Users need it."

"No," I told him. "Users ignore it."

He wasn't wrong about the effort. But I had the analytics to prove my point. And what we replaced it with changed everything.

The Problem Nobody Talks About

Here's the dirty secret about chatbots: they get 2-5% engagement. That's it.

We had a beautifully designed chat widget sitting in the bottom-right corner of our SaaS landing page. It could answer FAQs, route support tickets, even book demos. On paper, it was perfect.

In practice? 97% of visitors never clicked it.

The ones who did would type one question, get a mediocre canned response, and bounce. Support tickets kept piling up. Our "AI-powered" chatbot was basically a glorified FAQ page with a chat bubble skin.

I started digging into why people weren't engaging:

Typing is friction. On mobile especially, nobody wants to type out their question.
Canned responses feel robotic. Users can smell a decision tree from a mile away.
No real actions. The chatbot could tell you where to find pricing, but couldn't take you there.

What We Built Instead

We replaced the chatbot with a voice AI widget. Instead of typing, visitors just talk.

But here's where it gets interesting — this isn't just speech-to-text piped into a chatbot. The voice AI can actually do things on the page:

Navigate to any section or page ("Show me pricing" → it scrolls/routes there)
Fill out forms on behalf of the user ("Book a demo for Thursday" → fills the form)
Take DOM actions — click buttons, open modals, toggle features
Answer questions with full context awareness of what's on screen

It's like giving every visitor a personal assistant that lives inside your website.

The Tech Stack

For the devs reading this, here's what's under the hood:

React for the widget UI and DOM interaction layer
Supabase for real-time session management and analytics
Vertex AI (Gemini 2.5 Flash) for the LLM backbone — fast enough for conversational latency
Custom DOM Action Layer — this is the secret sauce. We built a middleware that maps LLM function calls to actual DOM manipulations. The AI doesn't just suggest actions, it executes them.

The function calling pipeline works like this:

User speaks → speech-to-text
LLM processes intent + page context
LLM returns structured function calls (navigate, fillForm, clickElement, etc.)
DOM action layer executes in real-time
Response is spoken back via text-to-speech

Total round-trip latency: under 500ms.

The Results

We've run this in production for about a month now. The numbers speak for themselves:

Metric	Chatbot (Before)	Voice AI (After)
Engagement Rate	3.1%	41.2%
Sessions	~200/month	500+ in first month
Tool Success Rate	N/A	99.2%
Conversion Lift	Baseline	+47%
Avg. Response Time	1.2s	<500ms

The engagement jump alone was staggering. People are curious about voice. They click the mic icon just to try it. And once they start talking, they stay.

The Multilingual Game-Changer

Here's something I didn't expect to matter as much as it did: multilingual support.

Our voice AI supports 50+ languages out of the box. That includes Hindi, Tamil, Bengali, Telugu, Marathi — basically every major Indian language.

Why does this matter? India has 500+ million internet users who don't speak English as their primary language. For SaaS products targeting the Indian market (or any multilingual market), a text chatbot in English is essentially useless for half your potential users.

Voice changes that equation completely. People speak naturally in their language, the AI understands and responds in kind, and it can still execute all the same DOM actions.

We saw our Indian user engagement go from near-zero to our fastest-growing segment.

What I Learned

Voice removes friction that you didn't know existed. Typing a question requires effort. Speaking is effortless.
Action > Information. A chatbot that tells you "Click the Pricing tab" is useless compared to an AI that just takes you there.
Multilingual isn't a nice-to-have anymore. If you're building for global users, voice is the fastest path to true accessibility.
Speed matters more than accuracy. A 95% accurate response in 400ms beats a 99% accurate response in 3 seconds. People forgive small mistakes but won't forgive lag.

Try It Yourself

We packaged this into a product called AnveVoice so other developers can add it to their sites without building everything from scratch.

5 minute setup. No credit card required. Just drop in a script tag.

👉 anvevoice.app

If you're still running a chatbot widget and wondering why engagement is low — try letting your users talk instead. The results might surprise you.

What's your experience with chatbots vs. voice on websites? Drop a comment — I'd love to hear what's worked (or hasn't) for you.

How Voice AI is Killing the Traditional Chatbot in 2026

Adarsh Kant — Wed, 18 Feb 2026 18:47:57 +0000

The chatbot era is over. Not because the technology failed — but because users voted with their behavior. They ignored it.

After a decade of "How can I help you?" pop-ups, businesses are waking up to a brutal truth: traditional chatbots don't convert. The average chatbot has an 87% abandonment rate. Users don't want to type into a tiny box. They want things done.

Enter voice AI for websites — a fundamentally different approach that doesn't just answer questions, but takes real actions on behalf of the user. This is the definitive chatbot alternative in 2026, and here's why the shift is inevitable.

The Chatbot Problem Nobody Talks About

Chatbots were supposed to revolutionize customer experience. Instead, they became the digital equivalent of an automated phone tree.

The data is damning:

95% drop-off rate before users complete a chatbot conversation
Average session: 22 seconds — barely enough to read the welcome message
74% of users immediately ask for a human agent
Only 11% resolution rate beyond basic FAQs

The core issue isn't intelligence — modern LLMs understand complex queries just fine. The problem is interaction design. Chatbots force text-based, turn-by-turn conversations when users just want to accomplish a task.

A user doesn't want to chat about booking an appointment. They want the appointment booked.

Why Text Interfaces Fail on Mobile

This compounds on mobile, where 70%+ of web traffic originates. Typing on a phone is slow, error-prone, and frustrating — especially for the 500 million+ mobile-first users in India who are more comfortable speaking than typing. For them, text chatbots are fundamentally the wrong interface.

Voice AI vs Chatbot: What's Actually Different?

When we compare voice AI vs chatbot, we're not just swapping text for speech. That would be a chatbot with a microphone. The real difference is what happens after the user speaks.

Traditional Chatbot: User types → bot returns text/link → user manually navigates, clicks, fills forms → friction → abandonment.

Website Voice Assistant (Action-Based): User speaks "Book me a demo for Thursday at 3pm" → AI extracts intent + parameters → AI takes real DOM actions (navigates to booking page, selects date, fills details) → user confirms with voice → done. Zero typing. Zero clicking.

This is the shift: from information retrieval to task execution. A voice AI for website doesn't just tell users what to do — it does it for them.

The Action Layer

What makes a modern website voice assistant a true chatbot alternative is the action layer:

Navigate pages — pricing, booking, contact, any section
Fill forms — names, emails, dates, preferences from voice
Click buttons — submit forms, add to cart, start trials
Read dynamic content — speak back prices, availability, status
Handle multi-step workflows — complete checkout in one voice command

This isn't theoretical. AnveVoice ships this today with a single line of embed code.

Why 2026 is the Tipping Point

Four forces are converging to make voice AI for websites the dominant interface:

LLMs got fast enough. Sub-500ms response times make voice AI feel like talking to a person — not waiting for a loading spinner.
Speech recognition handles real speech. Accents, code-switching (Hindi-English), background noise, incomplete sentences — all handled. Dealbreaker in 2023, table stakes in 2026.
Browser APIs enable direct action. Web Speech API + modern DOM manipulation lets voice AI interact with any website without backend integration.
Users expect it. Siri, Google Assistant, and Alexa trained a generation to speak to computers. Text chatbots now feel dated — like encountering a Flash animation in 2020.

The Business Case: Voice AI vs Chatbot ROI

Here's what businesses report after switching from chatbots to a website voice assistant:

Metric	Chatbot	Voice AI
Engagement rate	3-8%	18-35%
Task completion	11%	47%
Average session	22 sec	1 min 45 sec
Support ticket reduction	15%	52%
Conversion lift	Negligible	2-4x
User satisfaction	2.1/5	4.3/5

Sources: Internal AnveVoice data, Gartner CX Survey 2025, Drift State of Conversational Marketing

Industry Impact

Healthcare & Dental: "I need a cleaning next week" replaces a 4-step form. No-show rates drop with voice confirmations.

E-commerce: "Show me red sneakers under ₹3000 in size 10" → AI filters, displays, adds to cart. Average order value up 23%.

SaaS: "Start my free trial" triggers form fill automatically. Landing page → trial conversion increases 3x.

EdTech: "Enroll me in the Python course starting March" — enrollment friction drops to near-zero.

The Competitive Landscape

Here's the thing most people miss: almost every voice AI company is building for phone calls, not websites.

Bland.ai ($56M raised) — AI phone agents. Enterprise-focused, $0.09/min. Great for outbound calling — useless for website visitors.
Synthflow / Dialora — No-code voice agents for phone/telephony only. No website widget.
Voiceflow ($39M funded) — Conversational AI builder. Powerful but requires months of design work.
Ada.cx ($190M+ raised) — Enterprise-only, chat-first, inaccessible to SMBs.
Retell.ai (YC-backed) — Voice agent API for developers. Phone-focused, requires coding.

AnveVoice occupies a different space: an embeddable voice widget that lives on your website. No phone numbers. No telephony costs. No API integration. One line of code, live in 5 minutes, works in 22 languages.

Common Objections (And Why They're Wrong)

"Our users prefer typing." No — they prefer whatever gets the task done fastest. On mobile (70%+ of traffic), that's voice. And voice AI still accepts text as fallback.

"Voice doesn't work in noisy environments." Modern browser noise cancellation is remarkably good. Plus, text fallback mode exists.

"It's too expensive to switch." Most voice AI for website solutions deploy via a single <script> tag. No backend changes. AnveVoice has a free tier — test it in 5 minutes.

"What about privacy?" Voice data is processed in real-time, not stored by default. On-device processing where possible, all transmissions encrypted.

How to Make the Switch

Step 1: Audit your chatbot. Check engagement, completion, and escalation rates.

Step 2: Identify high-value actions. What do visitors come to do?

Step 3: Deploy. With AnveVoice, it's one line:

<script src="https://cdn.anvevoice.app/widget.js" data-key="YOUR_KEY"></script>

Step 4: Configure actions. Map voice intents to DOM actions. Visual builder included.

Step 5: Monitor. Track actions completed, not just messages sent.

The Chatbot Alternative 2026 is Voice

The voice AI vs chatbot debate is settled. Users want to speak, and they want things done. Traditional chatbots served their purpose, but their era has passed.

Voice AI for websites isn't an incremental improvement — it's a paradigm shift from conversation to action.

Ready to Kill Your Chatbot?

Try AnveVoice free → Deploy a website voice assistant in 5 minutes. One line of code. 22 languages. Real actions, not just answers.

No credit card. No sales call. Just paste the code and talk to your website.

Start free trial | Book a demo | Documentation

Originally published at anvevoice.app/blog/voice-ai-killing-chatbot-2026

Built by ANVE.AI — making every website voice-first. Follow us @ANVEAI for updates.

Why Voice AI Will Replace Chatbots on Indian Websites by 2027

Adarsh Kant — Wed, 18 Feb 2026 15:11:03 +0000

Indian businesses are at a crossroads. The chatbot era with its clunky text boxes and decision trees is ending. By 2027, voice AI will be the default interface on Indian websites.

The Problem With Chatbots in India

India has 900M+ internet users, but most of them do not want to type. India is a voice-first culture. We talk, we do not type. Yet every SaaS company, D2C brand, and healthcare platform still forces users into text chatbots.

The data backs this up:

70% of Indian internet users prefer voice over text
Average chatbot completion rate: 23%. Average voice completion rate: 68%
India vernacular internet users (400M+) struggle with English-only chatbots

What Voice AI Does Differently

Voice AI is not just speech-to-text plus chatbot. It is a fundamentally different interaction paradigm:

1. Natural Conversation Flow
Users speak naturally. No menus, no buttons. The AI understands intent, handles interruptions, and responds in real-time.

2. Multilingual by Default
A voice AI agent can switch between Hindi, English, Tamil, and Telugu mid-conversation. Try doing that with a decision-tree chatbot.

3. Emotional Intelligence
Voice carries tone. A frustrated customer sounds different from a curious one. Voice AI can detect sentiment and adjust responses.

The Technical Shift

Three converging technologies are enabling this:

Real-Time Speech Processing

Latency has dropped from 3-5 seconds to under 500ms. Modern WebRTC plus streaming ASR pipelines make this possible even on 3G connections.

Large Language Models

GPT-4, Claude, and open-source models like Llama have made it possible to generate contextually accurate, natural responses. Combined with RAG, voice agents can answer domain-specific questions accurately.

Cost Collapse

Running a voice AI agent costs about $0.02/minute. A human call center agent costs $0.15-0.30/minute. For SMBs handling 500+ monthly queries, the math is overwhelming.

Real-World Impact

We have deployed voice AI agents for Indian businesses across healthcare, D2C, and SaaS. The results:

Healthcare clinics: 24/7 appointment booking with 94% success rate
D2C brands: 3x increase in product queries converted to orders
SaaS companies: 60% reduction in support ticket volume

One dental clinic went from losing 40% of after-hours calls to capturing 95% without hiring a single person.

Why 2027?

Three inflection points converge by 2027:

5G rollout completion in Tier 2-3 cities enables reliable real-time voice
LLM costs drop 10x from current levels
Consumer expectation shift as Gen Z users already expect voice interfaces

The question is not whether voice AI replaces chatbots. It is whether your business adapts before your competitors do.

Getting Started

If you are a developer or founder, here is what to explore:

WebRTC + Streaming ASR: Look at Deepgram, AssemblyAI, or Whisper for real-time transcription
LLM Integration: Use function-calling with GPT-4 or Claude for structured responses
Text-to-Speech: ElevenLabs, Play.ht, or open-source alternatives like Coqui
Full-stack voice platforms: Solutions like AnveVoice handle the entire pipeline

The voice-first web is coming. India, with its oral culture and massive vernacular population, will lead this transition.

What is your experience with chatbots vs voice AI? Drop a comment below.

AI Impact Summit India Day 2 Recap: 5 Things Every Indian Founder Should Know

Adarsh Kant — Tue, 17 Feb 2026 11:32:31 +0000

February 17, 2026 | Bharat Mandapam, New Delhi

Day 2 of the India AI Impact Summit 2026 just wrapped, and if you're an Indian founder building anything remotely connected to artificial intelligence, today was the day that redefined the playing field. While Day 1 was about spectacle — PM Modi inaugurating the expo, 600+ startups showcasing, global leaders flying in — Day 2 got down to business. Applied AI, sovereign models, healthcare transformation, and a government that's putting serious money where its mouth is.

Here are the five things that matter most if you're building in this space.

1. Gnani.ai Just Proved India Can Build World-Class Voice AI

The biggest product launch of Day 2 wasn't from a Silicon Valley giant. It was from Bangalore-based Gnani.ai, which unveiled Inya VoiceOS — a 5-billion-parameter voice-to-voice foundational model built under the India AI Mission.

What makes this special isn't just the parameter count. It's the architecture. Inya VoiceOS operates directly in acoustic and semantic space, bypassing the traditional speech-to-text → LLM → text-to-speech pipeline that most voice AI systems rely on. CEO Ganesh Gopalan demonstrated real-time multilingual conversations that felt remarkably natural.

For founders building voice-first products — whether it's customer support automation, vernacular commerce, or conversational agents — this changes the infrastructure equation entirely. You no longer need to stitch together three separate models. A single end-to-end voice model, trained on Indian languages, built by an Indian company, running on Indian compute. That's a first.

The message from Gopalan was pointed: "Production-ready voice AI, not flashy demos, will define the winners." As someone building in the voice AI space with AnveVoice, I couldn't agree more. The companies that win won't be the ones with the prettiest demo — they'll be the ones whose voice agents actually work reliably at scale, in real Indian accents, across real Indian languages.

2. Sarvam AI Is Building India's Sovereign LLM Ecosystem — And It's Not Just Talk

If there was one name that kept coming up across every panel and hallway conversation, it was Sarvam AI. The Bengaluru-based startup, selected under the India AI Mission to build the country's first sovereign large language model ecosystem, used Day 2 to showcase its model family:

Sarvam-Large: For advanced reasoning and generation tasks
Sarvam-Small: For real-time interactive applications
Sarvam-Edge: For compact, on-device deployment

The company has been granted access to 4,000 GPUs for six months to train these models, and early benchmarks suggest Sarvam is already outperforming several global models on Indic language tasks.

Why does this matter for founders? Because sovereignty isn't just a political talking point — it's a business advantage. If you're building products for Indian users in Indian languages, you need models that understand the cultural context, linguistic nuance, and code-switching patterns that define how 1.4 billion people actually communicate. Sarvam isn't trying to be the next ChatGPT. It's trying to be something ChatGPT fundamentally can't be — deeply, natively Indian.

For startups building on top of LLMs, the implication is clear: you'll soon have a high-quality, India-optimised foundation model stack to build on, potentially at a fraction of the cost of licensing from OpenAI or Anthropic.

3. Healthcare AI Is Moving From Research Papers to Real Patients

Day 2 featured a dedicated track on AI in healthcare, and the announcements were substantive. Two government initiatives — SAHI (Smart AI for Health India) and BODH (Biomedical Open Data Hub) — are creating the data and infrastructure layer that Indian healthtech startups have been begging for.

Union Minister Anupriya Patel addressed the elephant in the room directly: "Fears that AI will replace doctors and clinicians are largely misplaced." The framing from the government is clear — AI as a force multiplier for India's overstretched healthcare system, not a replacement.

What's exciting for founders is the data access piece. India's healthcare AI challenge has never been about talent or algorithms — it's been about data. Patient records are fragmented across a million clinics with zero standardisation. BODH aims to create an anonymised, structured biomedical data commons that any verified Indian researcher or startup can build on.

If you're building diagnostic tools, telemedicine platforms, or even voice-based health assistants (imagine a nurse triage bot that speaks Bhojpuri), the infrastructure is finally catching up to the ambition.

4. $200 Billion in Data Centre Investment Is Reshaping India's AI Infrastructure

This was the number that made everyone sit up. Union IT Minister Ashwini Vaishnaw confirmed that India is looking at $200 billion in data centre investments over the next two years, positioning the country as a global AI compute hub.

Vaishnaw framed AI as the "fifth industrial revolution" and made the case that India's combination of talent, data (1.4 billion people generating massive datasets), and now compute infrastructure makes it uniquely positioned to be an AI superpower.

For founders, the compute story is the most practically important. Every AI startup I know has wrestled with GPU access, cloud costs, and latency issues from running inference on servers 8,000 miles away. More domestic data centres mean:

Lower latency for real-time AI applications (critical for voice and video)
Data residency compliance without performance penalties
Competitive pricing as supply increases and competition heats up among providers

Combined with the India AI Mission's compute allocation (like the 4,000 GPUs given to Sarvam), the government is systematically removing the infrastructure bottleneck that has held back Indian AI startups.

5. "Create in India" Is the New "Make in India" — And AI Founders Are the Target Audience

Perhaps the most significant policy signal from the summit was Minister Vaishnaw's announcement of the upcoming "Create in India" mission. While "Make in India" focused on manufacturing, "Create in India" is explicitly about building a talent pipeline and creating jobs in AI and emerging technology sectors.

The details are still thin, but the directional intent is clear:

Industry-oriented: Not just academic research, but production-ready AI
Employment-focused: Addressing the "will AI take our jobs?" anxiety head-on by creating new categories of AI-related employment
Future-ready talent pipeline: Training programs, skilling initiatives, and startup support

For founders, this is a signal that the regulatory and incentive environment is going to become increasingly founder-friendly for AI companies. If you're building AI tools, platforms, or products, you're not swimming against the current — the government wants you to succeed.

Nasscom AI Head Ankit Bose reinforced this on Day 2, stating that AI's rapid development "presents significant opportunities for Indian companies and will not adversely impact their businesses." That's the kind of institutional backing that gives investors confidence and founders runway.

What This Means If You're Building Right Now

If you're an Indian founder working in AI — whether it's voice AI, healthcare, fintech, education, or enterprise — the message from Day 2 is unambiguous: the ecosystem is finally aligning.

Sovereign models are coming. Compute is coming. Data commons are coming. Policy support is coming.

The founders who win will be the ones who:

Build for India first — not as an afterthought, but as a design principle
Choose production over demos — working systems that serve real users in real conditions
Leverage the sovereign stack — Sarvam for language, Gnani for voice, domestic compute for inference
Move fast — the window to establish category leadership is measured in months, not years

At AnveVoice, we've been building voice AI agents specifically for Indian businesses — dental clinics, healthcare providers, service businesses that need to handle calls in multiple languages, 24/7, without burning out their staff. Days like today validate the thesis: India needs its own voice AI infrastructure, built for its own realities.

The summit continues tomorrow with Day 3's Research Symposium, where IIIT Hyderabad leads a deep dive into AI research methodologies and emerging applications. We'll be watching closely.

The India AI Impact Summit 2026 runs from February 16-18 at Bharat Mandapam, New Delhi. Day 3 focuses on the Research Symposium, bridging academic AI research with real-world applications.

What India's AI Impact Summit Means for Developers Building AI Products

Adarsh Kant — Tue, 17 Feb 2026 08:36:47 +0000

The India AI Impact Summit 2026 is happening at Bharat Mandapam, New Delhi — and if you're a developer building AI products, several announcements directly affect your stack, your career, and your next side project.

Let me cut through the keynote fluff and talk about what actually matters for people who write code.

1. Sovereign LLMs You Can Actually Use

Sarvam AI and IIT Bombay's BharatGen are building large language models designed for Indian languages from the ground up. Not fine-tuned Llama. Not translated GPT. Purpose-built multilingual models.

Why this matters for devs:

API costs drop when you're not paying OpenAI per token for Hindi inference
Better accuracy for code-mixed text (Hindi-English, Tamil-English) than any multilingual adapter
Potential for self-hosted deployment — actual sovereignty over your inference pipeline

What to do: Watch for Sarvam's API access. If you're building anything serving Indian language users — chatbots, voice agents, content tools — test these against your current GPT/Claude/Gemini setup. The cost-performance tradeoff might surprise you.

2. SAHI & BODH — If You Build Healthcare AI, Read This

The government launched two frameworks:

SAHI (Strategy for AI in Healthcare for India) — governance framework for safe, ethical healthcare AI
BODH (Benchmarking Open Data Platform for Health AI) — privacy-preserving benchmarking, built by IIT Kanpur and the National Health Authority

For developers, this means:

Clear guidelines on what healthcare AI can and can't do
A benchmarking platform to test your models against standardized datasets
Privacy-preserving evaluation — you can benchmark without exposing patient data

If you're building health-tech, BODH is going to be the standard your models are measured against. Get familiar early.

3. "Create in India" — What It Actually Means for Dev Careers

IT Minister Vaishnaw announced a "Create in India" mission — government backing for building original technology, not just IT services. This includes AI talent pipelines.

Translation for developers:

More AI engineering roles at Indian companies (not just FAANG)
Government-funded AI research programs and grants
Startup ecosystem support = more interesting early-stage companies to join or build

The shift from "services" to "products" has been talked about for a decade. The government formally backing it with a named mission and budget is new.

4. $200B Data Centre Investment = Cheaper GPU Access

India wants $200 billion in data centre investments. For developers running inference workloads, this eventually means:

Indian cloud GPU providers competing on price
Lower latency for Indian users (no more routing through Singapore)
Potential for edge inference nodes in tier-2 cities

Right now, if you're training or serving models, you're probably on AWS Mumbai or Azure Central India. More data centres = more competition = better pricing. Simple economics.

5. The Open Source India AI Stack

Here's what's emerging as India's open-source AI ecosystem:

Foundation Models: Sarvam LLMs, BharatGen (more language-specific models coming)
Speech (STT): AI4Bharat's IndicWhisper, Vakyansh (improved code-switching models)
Speech (TTS): AI4Bharat's IndicTTS (regional accent support)
NLP: IndicNLP, IndicBERT (domain-specific legal, medical coming)
Benchmarks: IndicGLUE, BODH (healthcare-specific evals)
Datasets: Bhasha, IndicCorp (government-backed data programs)

This stack is maturing fast. Two years ago, building a Hindi voice assistant meant cobbling together Google Translate and hoping for the best. Today, there are purpose-built components for each layer.

6. The Multilingual Engineering Challenge (and Opportunity)

PM Modi trying Sarvam Kaze — AI glasses that respond in Indian languages — highlights the biggest unsolved engineering challenge in Indian AI: real multilingual support.

Not "we added Hindi translation." Real multilingual means:

Code-switching detection and handling (Hindi-English in one sentence)
Regional accent robustness in STT
Natural TTS in 22 languages, not robotic
NLU that understands cultural context, not just translated text

This is where Indian developers have a genuine edge. You understand the problem space intuitively. You code-switch daily. You know the difference between Mumbai Hindi and UP Hindi.

If you're looking for a hard, meaningful, career-defining technical challenge — multilingual AI is it. And the market is 1.4 billion people.

7. What to Build Right Now

Based on the summit signals, here's where developer opportunity is highest:

Voice AI applications: The Sarvam Kaze moment signals voice-first is the direction. Voice agents, voice search, voice commerce — all underbuilt for Indian languages. High demand, few good solutions.

Healthcare AI tools: SAHI + BODH create a clear framework. Build within it. Triage bots, appointment scheduling, patient follow-up systems. The regulatory clarity is an invitation.

Developer tools for Indian AI: The stack is fragmented. Whoever builds the "Vercel for Indian AI" — easy deployment, built-in multilingual support, pre-integrated with sovereign models — wins.

Privacy-preserving ML: BODH's privacy-preserving benchmarking approach will influence other sectors. Federated learning, differential privacy, secure inference — skills that are increasingly mandatory.

The Bottom Line

The AI Impact Summit's biggest message for developers: India is building its own AI stack, and there's room for you in it.

Not as a consumer of APIs. As a builder. The models are being created (Sarvam, BharatGen). The infrastructure is being funded ($200B). The governance is being established (SAHI, BODH). The market is massive (1.4B people, 22 languages).

What's missing is the application layer — the products that take all of this and solve real problems for real users. That's where developers come in.

The summit ends. The building doesn't.

If you're building AI for Indian users, I'd love to connect. I'm working on AnveVoice — multilingual voice AI agents. Drop a comment or reach out.

5 Ways Voice AI Beats Chatbots for Indian E-commerce Sites

Adarsh Kant — Mon, 16 Feb 2026 18:05:04 +0000

India's e-commerce market is projected to hit $200 billion by 2027. Yet most online stores still rely on text-based chatbots that feel robotic, frustrate mobile users, and completely ignore how Indians actually communicate online.

If you're building or running an Indian e-commerce site, here are five reasons voice AI isn't just better than chatbots — it's a fundamentally different experience for your customers.

1. Hindi-English Code-Switching Support

Here's how a real Indian shopper talks:

"Bhai, mujhe red colour ka shoes dikhao under 2000 wala"

That's not Hindi. That's not English. It's Hinglish — and it's how 350+ million Indians communicate daily. Traditional chatbots trained on clean English or formal Hindi choke on this.

Voice AI models built for the Indian market handle code-switching natively. At AnveVoice, we trained our speech recognition pipeline specifically on Hinglish patterns — mixing Hindi syntax with English nouns, brand names, and numbers.

The result? Your customer speaks naturally, and the AI actually understands. No forcing users to type in "proper" English. No frustrating "I didn't understand that" loops.

2. Hands-Free Mobile Browsing (India Is 95% Mobile)

Let's talk numbers:

95% of Indian internet users access the web primarily via mobile
Average screen size: 6.1 inches
Average browsing session: done while commuting, cooking, or multitasking

Typing on a tiny screen while navigating an e-commerce catalog with 10,000 SKUs is painful. Chatbots make this worse — they add a text input box on top of an already cluttered mobile UI.

Voice AI flips this entirely. Your customer just speaks:

"Show me cotton kurtis under 800 rupees in blue"

No typing. No scrolling through filter dropdowns. No pinching to zoom. The AI processes the request and navigates the catalog directly.

For mobile-first markets like India, voice isn't a nice-to-have. It's the natural interface.

3. Higher Engagement Rates (3x vs Chatbots)

Here's what the data shows across Indian e-commerce deployments:

Metric	Text Chatbot	Voice AI
Avg. session duration	45 seconds	2.5 minutes
Product discovery rate	12%	38%
Cart addition rate	3.2%	9.8%
Return visit rate	8%	24%

Why the massive difference? Voice is intimate. When someone talks to your site and it talks back in a natural voice, it creates a connection that text never will. Users explore more, ask follow-up questions, and stay longer.

The 3x engagement isn't magic — it's what happens when you remove friction from the discovery process.

4. Accessibility for Semi-Literate Users

This is the point most tech articles ignore, but it's arguably the most important for India.

India's literacy rate is 77%. But functional digital literacy — the ability to read, comprehend, and navigate a text-heavy e-commerce UI — is much lower. Millions of Indians who shop on Flipkart, Meesho, and JioMart struggle with:

Reading product descriptions
Understanding filter options
Typing search queries correctly
Navigating multi-step checkout flows

Voice AI makes your e-commerce site accessible to every potential customer, regardless of literacy level. A farmer in Madhya Pradesh can ask for "best fertilizer spray pump under 5000" in Hindi and get results — no reading required.

This isn't just good UX. It's market expansion. You're reaching customers that chatbots literally cannot serve.

5. Natural Product Search by Speaking

Text search on e-commerce sites is broken. Users type "red shoe" and get 4,000 results. They type "comfortable red shoe for office" and get zero results because the search engine doesn't understand intent.

Voice AI changes search from keyword matching to conversational understanding:

User: "I need a gift for my wife's birthday, she likes traditional jewelry, budget around 5000"

A chatbot would say: "Please search for jewelry in our catalog."

A voice AI agent would:

Understand the context (gift, wife, birthday, traditional, budget)
Filter the catalog intelligently
Present curated options
Ask follow-up questions: "Does she prefer gold-plated or silver?"

This is the difference between a search box and a shopping assistant.

Integration: It's Just One Script Tag

Adding voice AI to your site doesn't require rebuilding your frontend. Here's how you can integrate AnveVoice into any e-commerce site:

<!-- Add to your site's <head> or before </body> -->
<script src="https://cdn.anvevoice.app/widget.js"
  data-site-id="YOUR_SITE_ID"
  data-lang="hi-en"
  data-theme="auto">
</script>

<script>
  // Optional: Customize voice AI behavior
  AnveVoice.init({
    language: 'hi-en',        // Hindi-English code-switching
    voice: 'priya',            // Natural Indian female voice
    actions: {
      search: true,            // Enable voice product search
      navigate: true,          // Allow page navigation
      cart: true,              // Voice-driven add to cart
      checkout: false           // Disable for security
    },
    catalog: {
      selector: '.product-grid',
      priceField: 'data-price',
      nameField: 'data-name'
    }
  });
</script>

The widget is under 45KB gzipped, loads asynchronously, and doesn't impact your Core Web Vitals. It works with Shopify, WooCommerce, custom React/Next.js storefronts, and plain HTML sites.

The Bottom Line

Chatbots were built for a Western, desktop-first, English-only internet. India's e-commerce reality is different:

Multilingual (22 official languages + Hinglish)
Mobile-first (95%+ on phones)
Voice-native (users prefer speaking over typing)
Diverse literacy (not everyone reads fluently)

Voice AI isn't a chatbot upgrade. It's a completely different approach to customer interaction — one that's built for how Indians actually shop online.

If you're building for the Indian market, try AnveVoice free and see the difference voice makes.

I'm Adarsh, founder of AnveVoice. We're building voice AI that actually does things on websites — not just chat. Follow our journey on Twitter/X @ANVEAI.

Why We Built a Voice AI That Takes Real Actions on Websites (Not Just Chats)

Adarsh Kant — Mon, 16 Feb 2026 10:51:51 +0000

Every website chatbot in 2026 does the same thing: answers questions from a knowledge base. Some are better than others, but they all share one fundamental limitation — they can't DO anything on your site.

The Problem

A visitor lands on your e-commerce site and asks "Show me red shoes under ₹2000." A traditional chatbot responds with a text list. The visitor still has to navigate, filter, and find the products themselves.

That's not an AI assistant. That's a glorified FAQ.

What If AI Could Actually Act?

We built AnveVoice to solve this. Instead of just chatting, our voice AI agent:

Navigates to the right page when a user asks
Fills forms based on voice commands
Clicks buttons and triggers actions on the DOM
Guides visitors step-by-step through complex flows
Speaks back in natural voice (multilingual — Hindi + English)

It's the difference between "Here's a link to our pricing page" and actually taking the visitor there.

The Tech Behind It

Without going too deep:

Real-time DOM manipulation via a lightweight JS widget
Speech-to-text + LLM reasoning + text-to-speech pipeline
Action mapping: natural language → DOM selectors → browser events
Works on any website with a single script tag embed

Why India First

We're an Indian startup building for Indian businesses first. Here's why:

500M+ new internet users coming online prefer voice over typing
Hindi-English code-switching is natural — our AI handles it
₹999/mo pricing makes it accessible to SMBs (not $299/mo enterprise pricing)

Try It

Visit anvevoice.app — there's a free tier. Drop the script tag on your site and your visitors can start talking to your website in minutes.

I'm Adarsh, founder of AnveVoice. Building in public. Follow the journey on Twitter/X.