Key Details
At IIM Calcutta’s Lattice 2026 conclave, Digital India BHASHINI CEO Amitabh Nag outlined what is required to move language AI from experimentation to population-scale use.
Current reach: BHASHINI powers more than 800 government websites.
Usage: The platform has enabled over 9 billion cumulative AI inferences and processes more than 24 million daily.
Language coverage: It supports 36 Indian text languages, 23 Indian voice languages and 35 international languages.
Live application: BHASHINI’s Shrutlekh provided real-time transcription and translation during the address.
Development model: Community participation through BhashaDaan contributes language data and resources.
Policy status: The address presented implementation experience and priorities for language-AI adoption; it did not announce a new scheme, funding allocation or regulatory framework.
Language AI Must Address Three Connected Divides
Nag identified the digital, language and literacy divides as interrelated barriers to accessing online services. Interfaces designed primarily for literate users working in dominant languages can remain inaccessible even when connectivity is available.
Voice-first and multilingual tools can reduce this barrier, but India’s linguistic diversity requires systems to recognise variations in language, dialect, domain and context. Population-scale inclusion therefore depends on how reliably language technologies work in actual public-service settings, not simply on the number of languages nominally supported.
BHASHINI Is Moving from Platform Capacity to Live Use
The Digital India BHASHINI Division, under the Digital India Corporation and MeitY, provides shared language-AI infrastructure that government departments and other participants can incorporate into digital services.
The live use of Shrutlekh at the conclave illustrated one such application: converting speech into text and translating it in real time. BHASHINI’s deployment across more than 800 government websites indicates growing institutional use, while its daily inference volume provides a measure of activity across the platform.
These figures establish reach and usage, but do not by themselves show translation accuracy, user satisfaction or improvements in access to particular public services.
Scaling Problems Often Lie Beyond the AI Model
Drawing on BHASHINI’s development experience, Nag argued that some of the hardest problems concerned market design, incentives, organisational arrangements and ecosystem coordination, rather than the underlying AI model.
Moving from a prototype to a dependable service requires:
clearly defined uses;
closer interaction between research and product teams;
rapid testing and feedback;
representative language data;
partnerships that support deployment; and
incentives for institutions and developers to use shared infrastructure.
BhashaDaan, which invites communities to contribute language resources, reflects the importance of participation in building datasets for India’s linguistic diversity. Its value will depend on the quality, coverage and responsible governance of the data collected.
Shared Standards Can Expand the Platform’s Value
Nag encouraged developers and organisations to think beyond individual applications and consider the platforms and standards needed for wider participation. Common technical interfaces and reusable language services can allow multiple government departments, businesses and developers to build on the same underlying infrastructure.
This approach can create network effects: more contributors improve language resources, while more applications expand their practical value. It also makes interoperability, quality assurance and accountability important because weaknesses in a shared service may affect many downstream applications.
What Is an AI Inference?
An AI inference occurs when a trained model processes an input and produces an output. A speech-transcription request, translation or voice-language query may each generate one or more inferences.
Inference numbers indicate how frequently AI services are being used computationally. They are not equivalent to the number of individual users, completed transactions or successful public-service outcomes.
Policy Relevance
BHASHINI could make digital public services more accessible to people who face language or literacy barriers, but reach must be assessed through service performance as well as platform activity.
Government adoption: Ministries and states need suitable use cases, integration capacity and staff support to incorporate language AI into public-facing services.
Quality across languages: Testing should cover dialects, specialised terminology and real-world speech conditions, particularly for languages with fewer digital resources.
Accountability: Public deployments require clear responsibility for inaccurate translations, data protection, accessibility and routes for users to obtain human assistance.
Evidence of inclusion: Usage statistics should be complemented by measures such as accuracy, completion rates and whether multilingual tools help people access services independently.
Relevant Question for Policy Stakeholders: How should BHASHINI measure whether its language-AI services are improving access to public services across languages, regions and levels of literacy and not merely increasing platform usage?
Follow the Full Update Here: BHASHINI CEO Highlights the Institutional Requirements for Scaling Language AI