Electronics Era: How do you differentiate Devnagri AI from global language AI platforms, especially in the context of India’s linguistic diversity?
Nakul Kundra: The Global language AI platforms operate at scale across different geographical regions, while Devnagri AI concentrates on delivering deep language capabilities throughout Bharat. The linguistic diversity of our country extends beyond word translation into multiple languages. It requires translation of cultural elements together with regional-specific context and compliance requirements, and intent understanding. Devnagri establishes its unique position by developing specialised language models that use Bharatiya enterprise and government databases to create domain-specific models for sectors like BFSI and GovTech that demand precise and traceable results.
Devnagri offers multiple language functions through its infrastructure system, which enables voicechat, document and workflow systems to operate effortlessly across different language functions. It provides support for both on-premises and sovereign system installations and maintains regional language distinctions through various speech patterns, formal and informal communication styles.
Devnagri provides government agencies and businesses with a system that enables them to speak to all people across Bharat through accurate and compliant communications, which they can deliver at high volume.
Electronics Era: How critical is native-language AI in driving digital inclusion across sectors like electronics manufacturing, public services, fintech, and healthcare?
Nakul Kundra: Authentic AI in the native language is critical to genuine Indian digital inclusion. Humans must learn and use digital tools in the languages they speak every day. Most of this learning is driven by the non-English-speaking masses. They wish to consume content and services in their native Indian languages, mostly in rural or semi-urban sectors.
In the world of fintech and banking, English incorporated machines confuse the consumers about agreements, products and promote B2B exclusion. Penetration of native language AI increases usage, trust and lifetime loyalty. In industries, digital transition replaces field movements with online SOPs, fresher training, and quality management tools. Workers need simple, indigenous and relevant instructions to obey safety. Measures, minimise mistakes, and increase productivity.
Guidelines of Domestic Key Fact Statement (KFS) by RBI require the loan lenders to explain the same about themselves, the borrowed loan, the charges, and deviations in a simple language. Our firm uses AI for voice robots and an indigenous language translator to enhance information dissemination, and not only 100% compliance. For UPI apps and digital banking operations, optimal efficiency happens only when all transactions are conducted in local regional languages.
The Indian digital divide is beyond connectivity, it’s local inclusion through indigenous language by embedded AI, and the technology becomes meaningful, accessible, unstoppable, enabling true inclusion, and transforming one to two.
Electronics Era: Electronics manufacturing is becoming increasingly data-driven. How can language AI improve efficiency in documentation, training, quality control, and customer support?
Nakul Kundra: As electronics manufacturing processes become increasingly data-centric, language AI is the connective tissue, people and processes together, dominating on the shop floor and value-chain interactions.
In documentation, language AI can automatically localise SOPs, work instructions, quality logs and audit reports, providing a unified, standardised view with reduced translation errors, turnaround and cost- crucial in regulated production environments.
For training, native language conversational AI helps rapidly integrate shop floor workers and keep them up to speed: voice and chat-enabled assistants can guide safety practices, maintenance routines and quality regulations in employees language of choice, improving safety and yield.
In quality control, language AI can ingest unstructured voice, inspection and incident inputs and turn them into structured views to reveal recurring quality problem areas. For customer and vendor engagement, multilingual language AI removes language barrier by accurately translating communications, powering more responsive service in applications ranging from order tracking to warranty explanation.
Language AI makes data work smarter for every player, strengthening compliance, customer service, and predictive insights at scale.
Electronics Era: Many enterprises remain cautious about deploying AI due to concerns around accuracy, data privacy, and bias. How does Devnagri address these challenges?
Nakul Kundra: Enterprise caution around AI is well justified. Accuracy, data privacy and bias cannot be an afterthought. They have to be enshrined in the design from the get-go. That principle defines the core of how we construct Devnagri.
Accuracy arises from our multilingual foundation models, which are finetuned with 700k+ domain-specific local datasets and are aligned across language, voice, tone, and industry-vertical terminology. These datasets are anonymised and consent-managed, while our model quality is incrementally optimised through auto-evaluation signals including self-consistency checks, semantic coherence scoring after use feedback.
Data security and privacy are limited to enterprise control. Devnagri enables private cloud and on-premise deployments, with least privilege access control, 24/7 audit logs and data residency taking into account the India Digital Personal Data Protection Act as well as enterprise standards.
Bias and cultural misinterpretation is driven at the data and checkpoint evaluation levels. Responding to highly localised datasets of languages empowers our models to represent real-world language, not watered-down assumptions. Our perspective is no AI is responsible if its architecture is not responsible. It is with governance, evaluation and human controls incorporated into the design that an enterprise can confidently grow AI.
Electronics Era: How important is human-in-the-loop AI when dealing with critical industrial and technical content?
Nakul Kundra: For industrial, technical and mission-critical data, human-in-the-loop AI is an absolute necessity. Industries such as Manufacturing, BFSI, Healthcare and Public Infrastructure rely upon communication at risk of operational, financial or safety failure. A single negative instruction, mistranslated safety note or misleading compliance clause can cause outages or endanger lives. Many indisputably intelligent purely AI or generic ML models were never designed to take that level of responsibility.
At Devnagri, AI is a tool to assist humans, not replace them. Human-in-the-loop enabled workflows provide:
Root-cause analysis and quality improvements from our domain experts Reliance and compliance for essential policies, safety documents and customer communications Models that evolve with real-world correction and edge cases Particularly for Indian languages, this is very important. Variations in expression, meaning and tone can be subtle to the machine but very significant in communication. Human input minimises avoidable errors and unintended consequences.
Autonomous AI takes trust; human-in-the-loop AI takes control. It allows organisations to confidently upscale their language workflows with that extra assurance of reliability, accuracy and trust.
Electronics Era: What are the limitations of large global language models when applied to Indian languages and industry-specific use cases?
Nakul Kundra: Large global language models are super-scale, but have architectural constraints when used for Indian languages and industry-specific applications.
First, most global models are developed mainly on the English language and internet-scale data, so Indian languages are often not properly represented, and often handled as translations, rather than first-class languages. As such, there are deficiencies in dialects, code-mixed speech, regional nuance, and culturally-specific expressions, which are key for real interactions in Bharat.
Second, the generic design of global models means that they have a limited understanding of industry-specific context, such as Kashushk, manufacturing process, healthcare domains, or government workflows. In such regulated or technical domains, errors, hallucinations, or generic performance may deliver pretty sounding language that may be operationally inaccurate or non-compliant.
Third, global models typically operate in shared, cloud-first contexts, meaning data security, sovereignty and auditability issues. Most Indian companies and government organisations require on-prem or controlled deployments, and traceability functionalities that mainline global models do not offer.
Fourth, the issues of bias and exclusion also affect global models. Black-box global training data that emphasises urban-centric, anglo-Anglophone data may tend to alienate regional users and lead them to the wrong conclusions.
While super-scale large language models are a very good beginning point, they are not sufficient alone for band-scale, mission-critical Indian use cases. This is where a native-Indian language model, built by India, for India, with cultural, compliance and control being built in, becomes inevitable.
Electronics Era: India is emerging as a global AI talent hub. How is Devnagri leveraging local research and engineering talent?
Nakul Kundra: India is not just a cost-effective AI talent pool anymore,it is becoming a source of original AI thinking, especially in areas like language, scale, and real-world constraints. At Devnagri, we are deeply intentional about leveraging this local talent advantage.
We design our core technology in India, working with teams that understand Bharatiya languages, dialects, and user behaviour firsthand. This positioning helps our researchers & engineers to solve the large-scale problems that are not visible to global teams, like code-mixed speech, regional tone, low-resource languages, enterprise accuracy, and regulation of the above.
From a design perspective, we discipline ourselves to traverse research and production engineering and not work in isolation on blue-sky models. Our models are in the pan, monitored and improving on the live with banks, governments, etc.
So the most important thing we do is to enable our engineers to think beyond benchmarks focussing on impact at a Bharat scale where scores are irrelevant, but adoption and trust matter. And that is how India will make unique & in-market language-specific AI innovation, and that is exactly what Devnagri is built on.
Electronics Era: How does Devnagri plan to scale globally while remaining deeply rooted in India’s linguistic and cultural context?
Nakul Kundra: Devnagri’s approach to global scale is deliberate and layered. We scale the platform globally while keeping the linguistic and cultural core deeply rooted in Bharat.
Our foundation is a multilayered language-first, cultural contextSensitive structure. We do not treat Indian languages as English translations, but as native language systems with their own rules, tone, and context of use. This discipline forms a solid core that can then be extended into other multilingual markets with similar nuances.
Our global expansion strategy is partner-led, starting in the GCC by considering area factors, multilingual crowds, overly-regulated sectors, and compliance standards that map India seamlessly. Instead of transplanting our global-first models, we downshift expanding the playbook domain &trained algorithms, local data partnerships, human-in-the-loop processes, carefully dialling out deployments.
Notably, Indianness remains a differentiator &strength-we care about its diversity, regulation &inefficiencies from the get-go, building products that are high volume, high accuracy, and high flexibility from the start. Once the system works with Indian data, languages, and enterprise needs, we know it is ready to go anywhere.
In essence, Devnagri scales globally by exporting capability and methodology, while staying anchored to India’s linguistic and cultural intelligence. That balance allows us to be globally relevant without losing local authenticity.
Electronics Era: How do you see language AI influencing India’s economic growth over the next decade?
Nakul Kundra: Over the next decade, language AI will be one of the most powerful accelerators of India’s economic growth because it will bring the next wave of citizens, workers, and small businesses into the formal digital economy.
India’s economic upside is tightly linked to participation. Today, a large part of Bharat still faces friction in adopting digital services such as banking, healthcare, education, governance, and commerce because interfaces and processes are often not designed in the user’s native language. Language AI removes that friction by making technology understandable, voice-enabled, and culturally familiar, which directly increases adoption and trust.
Practically, we expect language AI to propel growth in 3 fundamental ways:
Productivity at scale: Companies will automate India-relevant support, docs, training, and compliance in all languages, lowering costs while raising quality.
Inclusion in financial and digital services: vernacular-down strategy onboarding, consent, and servicing will enable credit, insurance and saving-including for tier 2/3 & rural Indians.
Improved delivery of public services: Govt can communicate schemes, gather perspectives, and redress complaints in native languages- improving targeting and reducing leakage, resulting in higher impact for the same spending.
Most importantly, language AI will turn language into an infrastructure layer, enabling hundreds of millions to learn, transact and work faster. Bharat that can participate digitally in its own languages, is not only growing, but it will also be larger, more resilient, and more equitable.
Electronics Era: Why should enterprises view language as infrastructure rather than a feature?
Nakul Kundra: Enterprises need to think about language as infrastructure because it drives adoption, trust, and scaling, the same way that payments, identity or connectivity do. It is a feature when it remains cosmetic and isolated. It becomes infrastructure when it becomes an integral part of operations and growth.
In a country like Bharat, language is not a peripheral UX choice but a strategic one because it can make or break the ability of customers or employees to make sense, comply and complete. Like identity and payments, it streamlines every core workflow: onboarding, consent, support, training, collections, and grievance resolution. When language can be bolted on late in the process, enterprises face higher costs, lower adoption and elevated business risk.
As infrastructure, language has to be:
Consistent across touchpoints, voice, chat, documents or apps Integrated deep into workflows, not just on the edges for translation or transliteration
Managed, governed and audited, especially within the regulated industry Reusable and scalable for different products or markets every time enterprises it from scratch.
Enterprises that design language as infrastructure have a distinct money-making advantage of faster onboarding, enhanced compliance, reduced support burden and extensive market scope. Whereas simple features offer limited impact and ongoing headwinds.
In a multilingual economy, language is not a layer you add—it’s the foundation you build on.
