India ‘Talks’ The AI Walk

Welcome to The AI Shift by Inc42, our all-new newsletter that delves deep into the world of artificial intelligence, LLMs, big tech giants and the major trends sweeping the Indian startup and tech ecosystem. Here’s the second edition; do send us your feedback and suggestions so we can improve as we go along!

Indians did not learn to engage with technology because of AI. We have been doing it for years. Long before smartphones, dashboards, and mobile apps were the norm, we engaged with technology through customer support, IVRS, and call centres. Speaking, not typing, was how things really worked back then.

However, as technology grew its influence, the nation’s digital ecosystem began to be built around text. For a country like India, with several language barriers and varying literacy levels, this is a problem — and it did take its toll. Unlike more advanced nations, tech adoption became a challenge, as text-heavy, English-first interfaces made it difficult for millions of users to understand how to interact with digital systems.

Fast forward to 2026, India’s tryst with tech stands at an interesting point, especially when voice and AI have come together to make digital systems more accessible and better aligned with how people naturally communicate.

From government initiatives like Bhashini, which aims to reduce language barriers through voice, to AI startups building Indic LLMs and speech models that understand the nuances of Indian languages, this shift cannot be more profound.

But as voice moves from habit to infrastructure, the bigger question is who is building it, where it is being applied, and what problems it is actually solving. Let’s find out in this week’s edition of The AI Shift.

Currently, the global voice AI startup landscape is being dominated by names like ElevenLabs, Deepgram, Wispr Flow and AssemblyAI, which are enabling voice-driven workflows by building text-to-speech (TTS) and speech-to-text (STT) models and voice cloning.

The country’s voice AI ecosystem is evolving with startups such as Gnani.ai, which has launched Vachana STT, a foundational Indic speech-to-text model; Sarvam, which is building sovereign multilingual voice and LLM infrastructure; and Smallest.ai, which is focussed on delivering responsive, scalable TTS and voice systems. CoRover.ai is powering conversational AI bots, while Oriserve is building enterprise voice AI agents for a range of business use cases.

What this really means is that voice is being designed as the primary interface rather than an accessibility layer, especially for high-frequency, real-world interactions.

For Akshat Mandloi, the founder and CTO of Smallest.ai, voice feels inevitable because it mirrors how humans naturally communicate.

This shift is already reshaping how commerce happens.

As Mathangi Sri Ramachandran, cofounder and CEO of YuVerse, puts it, “Voice is a revolution at this point. India is getting completely into conversational commerce, and that’s becoming mainstream commerce.”

She pointed out that voice is no longer restricted to support or reminders, and a lot of business tasks – from reminding the customer to pay to exposing them to a product, upsell, and cross-sell – are already happening on voice.

For Anurag Jain, founder and CEO of Oriserve, voice clearly outperforms apps and text in customer-facing situations where discovery, urgency, or emotion is involved. “This is especially true in areas like collections, insurance servicing, and support,” he said.

He added that even digitally fluent users revert to conversation when friction rises. While tier I users are comfortable with apps due to muscle memory, they prefer calling when exploring something new because apps create a high cognitive load, leading to drop-offs.

However, what surprises Jain is how people frequently switch languages mid-sentence, rely on context rather than structure, and expect systems to keep up. This is precisely where the next big opportunity lies.

Besides, voice AI is lowering the cost of building tech for India’s fragmented, multilingual market. Traditionally, startups either limited themselves to a few regions or invested heavily in localisation, translation, and human support teams.

Speech-first systems have flipped that equation, and this matters deeply in a country where users may be digitally active but not comfortable navigating complex interfaces, especially in tier II and III markets.

As voice moves from interface to system, how it is built begins to matter as much as where it is used. Infrastructure control is emerging as a key advantage. Jain noted that Oriserve deliberately built its own stack to avoid external constraints.

The company has developed indigenous speech models purpose-built for Indian languages and dialects, along with patent-pending algorithms that can train on new languages within weeks using minimal data.

“Because we control the full stack, new use cases launch much faster,” he said, adding that BFSI and telecom have benefited the most, particularly where high-volume interactions demand instant scalability without constant hiring.

Voice-led systems are also expanding beyond traditional call-centre scenarios into physical and operational environments. Mathangi Sri Ramachandran of YuVerse shared an example where voice is used to troubleshoot vending machines during on-ground operations. She added that voice is going to occupy a lot of this commercial transaction space in India.

From an investor perspective, voice is now crossing a critical threshold. According to Arjun Malhotra, general partner at Good Capital, “Voice AI in 2026 appears to be transitioning from novelty to infrastructure.”

Smallest.ai’s strategy reflects this infrastructure mindset, for instance.

Mandloi said the company was built around a contrarian insight that large, general-purpose models are not always necessary. “Why do we need such large models to solve everything?” he asked. “To solve very specific business use cases, you don’t need very large models.”

That thinking led Smallest.ai to focus on real-time voice. “Not a lot of players were focussing on real-time conversations. We saw a gap there. Latency was the key constraint. Smaller models can achieve much smaller latencies, making them better suited for live conversations at scale,” Mandloi said.

Malhotra, meanwhile, pointed to the steady maturation of enabling guardrails. UPI 123PAY has normalised IVRS-based payments, while WhatsApp Business Calling API has unlocked VoIP-driven interactions at scale. Together, these allow users to transact, manage, and consume services via voice in ways that were not practical three years ago.

The more profound shift, however, is technical.

Unlike traditional natural language processing (NLP) systems that relied on rigid menus and structured inputs, LLM-based systems can interpret messy, interrupt-driven conversations, extract intent, entities, and outcomes, and keep conversations flowing naturally.

This capability unlocks an entirely new category of software, where speech is no longer perishable but becomes data.

Within this evolving landscape, Good Capital sees two distinct archetypes of voice AI startups emerging.

The first focusses on voice as data and orchestration — capturing raw speech and converting it into structured insights and automated workflows. These systems turn conversations into operational intelligence, enabling decision-making that was previously invisible to software.

The second archetype builds human-like, multilingual voice interfaces that can operate apps, manage SMEs, and assist businesses through natural spoken instructions. These systems move beyond assistants into active execution, helping businesses run day-to-day operations using speech as the primary control layer.

Together, these signals indicate that voice AI in India is evolving beyond assistants and entering the domain of workflow automation, decision support, and operational intelligence.

Voice is also becoming a key distribution layer across BFSI collections and renewals, conversational commerce and upselling, onboarding and activation, complex enterprise workflows, field operations, SME task management, and language-native access.

For founders and CXOs, the strategic insight is becoming clearer. Voice is not a UI add-on or a support feature. It is a distribution layer embedded deep inside workflows.

This is particularly powerful in India’s service-heavy economy, where many workflows are still conversational rather than dashboard-driven. According to Good Capital, voice-based tools work because they fit how SMEs, field teams and service businesses already operate, instead of forcing them to change their behaviour.

Voice AI systems are already helping businesses detect missed upsell opportunities, guide users through complex onboarding flows, and convert conversations into operational indicators. This dual value, productivity through automation and intelligence via structured data, is what gives voice its staying power.

Despite rapid model improvements, multiple experts agree that technology alone will not decide winners. Jain believes long-term differentiation will come from domain depth and thoughtful human involvement.

He added that AI doesn’t replace human creativity, it learns from it. Therefore, the real opportunity lies in using AI to eliminate non-productive human work while keeping humans in the loop where judgment and nuance matter.

This view reflects what is already happening on the ground.

Voice bots are handling high-volume, repetitive interactions, while humans step in for negotiation, exception handling, and trust-heavy moments. In several cases, voice has become the first interface, with humans acting as escalation layers rather than default operators.

Voice AI white spaces in India are emerging where friction, urgency, and language complexity break app experiences. The most potent opportunities may lie in voice-first resolution for stressed users, code-mix native speech stacks for ‘Hinglish’, and regional switching.

Many use cases like collections, renewals, and field-agent copilots could work best in BFSI, insurance, telecom, ecommerce, healthcare, logistics, and for any SMEs where calls dominate workflows.

Despite strong tailwinds, the sector faces constraints. Malhotra cautioned that unit economics remains a real challenge. Streaming automatic speech recognition, LLM reasoning, and TTS together cost $0.07–$0.15 per minute, which can be expensive for early stage startups without scale.

Localisation remains another hurdle. India’s linguistic landscape involves Hinglish, regional languages, cultural nuance, interruptions, and incomplete sentences. Global models still struggle to handle these patterns, and locally tuned datasets remain insufficient. Infrastructure and UX challenges also persist, particularly around mobile voice navigation and integration with existing enterprise systems.

Behavioural and regulatory considerations, including consent, privacy, and workflow integration, add further friction, especially in passive recording or analytics-driven use cases.

Voice removes friction, builds trust, and aligns with how Indians naturally communicate. It converts conversations into data, workflows into automation, and interfaces into invisible infrastructure.

As voice AI matures, India is shaping a category where technology finally meets users on their terms. And for startups building the next wave of digital products, voice may become the most powerful distribution unlock of this decade.

Viral AI moments are commonplace these days, but every once in a while, a weekend experiment snowballs into something much bigger. That’s exactly what happened when Pankaj Tanwar, a senior software engineer at InMobi, decided to hack his motorcycle safety helmet out of sheer frustration.

Tanwar shared that he was ‘tired of stupid people on the road’, so he turned his helmet into an AI-powered traffic policing tool. While riding, an AI agent runs in near real time, flags traffic violations, captures proof with number plates and location data, and sends it directly to the police.

What followed was unexpected. The post went viral, and the Bengaluru City police reached out to him on X. Soon after, his DMs were flooded with messages from founders, industry leaders, and even other state police officials.

Tanwar said he has already seen early inbound investor interest, interviews with national TV channels, coverage across newspapers, radio, and digital media, and thousands of messages from people inspired to build something themselves.

The helmet is still ‘hacky and early stage’, but Tanwar is now actively working on a roadmap. Sometimes, all it takes is a weekend build and the internet’s attention to turn frustration into momentum.

As India moves closer to full enforcement of the Digital Personal Data Protection (DPDP) Act, enterprises are grappling with a new reality. Managing sensitive data is no longer a one-time audit exercise, but an ongoing operational challenge that cuts across systems, vendors, and workflows.

Founded in 2024 by Shashank Karincheti and Amit Kumar, Bengaluru-based Redacto is building an AI-driven privacy infrastructure platform designed for continuous compliance. The startup focusses on helping enterprises move away from fragmented, manual privacy processes toward real-time data governance.

Redacto’s modular suite includes products such as Privacy Engine, ConsentFlow, VendorShield, and TrustCentre, which automate personal data discovery, consent management, third-party risk monitoring, and regulatory reporting. The platform continuously scans and classifies data, maps it across internal systems, and triggers alerts when consent or vendor-risk thresholds are breached.

Targeting BFSI, fintech, and large enterprises, Redacto combines GenAI-led automation with an India-ready, modular privacy stack, which enables organisations to meet regulatory requirements without disrupting existing workflows.

With India’s privacy-management software market projected to grow from $87 Mn in 2024 to $334 Mn by 2030, Redacto is betting that continuous, AI-powered compliance will become a core enterprise requirement rather than an afterthought.

What prompts and hacks are CTOs, CEOs and cofounders using these days to streamline their work? Here’s Ashutosh Prakash Singh, cofounder and CEO of RevRag.AI, with a prompt he uses to bring focus and structure to strategic decision-making as the company scales: “You are my chief of staff. Based on my company’s current stage, customers, and market, list the top five strategic priorities I should focus on in the next 90 days. For each priority, explain why it matters now, what to deprioritise, and what success looks like.”

Editor’s Note: Some prompts may need to be adjusted by users for best results or may not work as intended for certain users.