Key Points
- BHASHINI platform now processes over 1.5 crore AI requests daily across 36 Indian languages
- System powers more than 500 government websites with sub-second response times
- Platform operates on India's first fully sovereign AI cloud independent of foreign providers
The government’s BHASHINI language technology platform has crossed 600 crore total requests while adding new open-source AI models, making multilingual digital services accessible to Indians who do not speak English or Hindi.
The Digital India BHASHINI Division (DIBD), which operates under MeitY, announced integration of Sarvam’s open-source AI models into the platform. The addition strengthens BHASHINI’s ability to serve citizens in their native languages, from Tamil to Santali, whether they are filing government applications or accessing healthcare information.
For a farmer in rural Odisha seeking crop insurance details or a senior citizen in Kerala checking pension status, the platform translates government services into languages they understand. The system now handles over 1.5 crore requests daily, with response times under one second.
BHASHINI Platform Runs on Sovereign Cloud Infrastructure
BHASHINI operates on what the government calls India’s first vendor and cloud-agnostic AI Sovereign Cloud. This means all data processing happens on servers controlled entirely by Indian authorities, without dependence on foreign cloud providers such as Amazon Web Services or Microsoft Azure.
The platform manages over 350 AI models, which are specialised software programmes trained to understand and generate human language. These models power services across more than 500 government websites.
The infrastructure uses NVIDIA’s cloud computing framework to process requests. Total requests have crossed 600 crore since launch, according to official figures.
Translation Without Hindi as Middleman
One technical advancement addresses a longstanding problem in Indian language technology. Earlier systems translated, say, Tamil to Bengali by first converting Tamil to Hindi, then Hindi to Bengali. This two-step process introduced errors and delays.
BHASHINI’s neural machine translation, a method where AI learns language patterns from large datasets rather than following fixed grammar rules, now translates directly between Indian languages. Tamil converts to Bengali without the Hindi detour.
The speech recognition component, which converts spoken words into text, has been trained on publicly available datasets. This allows external researchers to verify and improve the system’s accuracy.
Over 20 Language Services Across 36 Languages
The platform offers more than 20 specialised services. These include automatic language detection, which identifies what language someone is speaking, and speaker diarisation, which separates different voices in a recording.
Text services cover 36 Indian languages, while voice services support 23 languages. The system also handles tribal dialects and over 35 international languages for documents involving foreign text.
Text-to-speech conversion, which generates spoken audio from written text, uses multiple AI components working together. A denoiser removes background noise from recordings, while normalisation tools handle the conversion between spoken and written forms, such as reading ‘₹500’ aloud as ‘five hundred rupees’.
What the Platform Enables
Amitabh Nag, CEO, Digital India BHASHINI Division, said the platform provides a complete system where AI models, computing infrastructure and applications work together.
“The integration of advanced open models and large-scale inference capabilities enables developers, institutions, and innovators to build, deploy, and scale AI solutions that are reliable, accessible, and designed for real-world use at population scale,” Nag stated.
Current applications include multilingual chatbots for citizen services, government portals that work in local languages and digital public service tools. Developers and startups can access the platform’s capabilities through application programming interfaces, which are standardised methods for different software programmes to communicate with each other.
The government has positioned BHASHINI as digital public infrastructure, similar to UPI for payments or Aadhaar for identity verification.
Your Questions, Answered
What is BHASHINI and what does it do?
BHASHINI is the government's AI-powered language technology platform that translates text and speech across 36 Indian languages. It powers over 500 government websites, helping citizens access services in their native language rather than English or Hindi.
How many requests does BHASHINI process daily?
The platform processes over 1.5 crore requests daily, with total requests crossing 600 crore since launch. Response times remain under one second.
What is a sovereign AI cloud?
A sovereign AI cloud means all data processing happens on servers controlled entirely by Indian authorities, without dependence on foreign providers like Amazon or Microsoft. This ensures Indian citizen data stays within Indian jurisdiction.
What are the Sarvam models added to BHASHINI?
Sarvam models are open-source AI models designed for Indian languages. Their integration allows developers, government bodies and businesses to build multilingual applications using publicly auditable technology.
