AIKosha provides access to multiple datasets and models to develop artificial intelligence (AI) applications. The compute portal allows researchers, academics, and startups to request access to powerful computing resources made available through more than 14,000 graphics processing units (GPUs) from 10 companies at subsidised cost.
On the AIKosha portal, the ministry outlined eight use cases to illustrate how the datasets and models can be used to enhance public services. For instance, AI voice assistants for rural citizens could be built using datasets of voice samples from Indian languages and dialects. These assistants could help people to apply for government schemes and use voice verification to prevent duplicate applications and fraudulent subsidy claims.
Also read | U.P. social welfare dept to use AI for better implementation of public utility schemes
The AI Compute Portal, which was launched on Wednesday, will give researchers, academics, and startups access to 14,517 empanelled GPUs. Until now, requests for access to the computing resources were being sent to the ministry of electronics and information technology (MeitY) through emails. These requests will now be processed through the portal, which will become fully operational within days.
The IndiaAI Mission was launched on March 7, 2024, with a budget of ₹10,371.92 crore. Of this amount, 44% ( ₹4,563.36 crore) is earmarked to provide computing capacity through 10,000 GPUs, which represents the largest of the mission’s seven pillars.
“The teams that bid for compute capacity have said that they will add more GPUs every quarter,” IT minister Ashwini Vaishnaw said.
Also read | How are Trump’s AI plans and China’s DeepSeek shaping Paris summit? Details
Sunil Gupta, CEO and co-founder of Yotta, said that the company, which is providing 9,216 GPUs (about 63% of all approved GPUs in the first phase), will also apply in the second bidding round that closes on April 30.
“Our progress for building India’s foundation models is advancing exceedingly well. We have received 67 applications,” Vaishnaw said. Twenty-two applications are for developments of large language models while others are for development of small language models. “A significant part of the computing capacity will be dedicated to developing our own foundation models,” he added, referring to the large AI systems that power many applications.
Also read | Kerala govt to develop state-owned AI engine model for schools
IT secretary S Krishnan noted that unlike other countries, where most investments in AI had been made by private companies, in India, the government has invested in a “clear, focused program to implement nationwide adoption of AI through seven dedicated pillars.” He said that both the government and private sector will continue adding to the common computing capacity “at a cost that is significantly lower than anywhere else in the world.”
The lowest bids for the different GPUs were already on average more than 40% below the market rate. For researchers, students, and startups, the government will further subsidise access by up to 40%.
Repository of databases
At present, AIKosha hosts 315 datasets and 84 models from 12 organisations, along with multiple tools. Vaishnaw said that the platform will only have non-personal data. These datasets include those from Digital India Bhashini division, where one dataset contains 1,684 hours of labelled speech data across 12 Indian languages, which could help develop tools for automatic speech recognition.
Another dataset from the Telangana government contains transaction data from the state’s fair price shops.
Abhishek Singh, the CEO of IndiaAI, said that AIKosha has been launched as an MVP (minimum viable product) with a limited number of datasets. “The platform exists, but it will need to be enriched,” he said. To develop applications based on the models and datasets, the division is gathering problem statements from different ministries and state governments to present them as challenges for researchers, startups, and industry in the next round.
Also read | Centre eyes domestic AI models as ‘cost-efficient’ DeepSeek disrupts tech world
To launch the platform, Meity prioritised datasets that researchers had requested in their applications under different IndiaAI Mission pillars, an official told HT on condition of anonymity.
The ministry is not transferring all datasets from the existing Open Government Data (OGD) repository, as AIKosha must contain AI-ready datasets. “We are selecting what is relevant and AI-ready in terms of data quality, standardisation, and other parameters,” the official said.
Meity is also training other ministries and state governments on how to clean their datasets and use them to create applications for public services.
Models available on AIKosha include multiple text-to-speech and translation models for Indian languages from Meity’s Bhashini Division.
Private companies such as Sarvam AI and Ola Krutrim have also uploaded their models to AIKosha. For instance, Sarvam AI’s Shuka can be used to develop audio-based question-and-answer tools in Indian languages. Krutrim’s Chitrarth can generate images from text descriptions in ten Indian languages in addition to English.
To download datasets, individual users must first sign up for DigiLocker using their mobile number. Organisations must register via Meity’s Entity Locker using Aadhaar. This approach aims to ensure platform security and the appropriateness of the datasets. “This is a government platform. What if tomorrow someone uploads a dataset with inappropriate images? We need to be accountable as well,” the official explained. The platform allows contributors to set access controls for the datasets and models they upload.
On AIKosha, Meity has also outlined eight use cases demonstrating how different models and datasets on the platform can provide public services — such as the voice assistants for rural citizens to help understand dialects. This would require datasets from Bhashini and Project Vaani (covering 54 Indian languages and dialects), data from government schemes like E-Shram and MNREGA, and information on regional speech patterns.
Thirty teams were awarded under the IndiaAI Application Development Initiative for developing AI-based solutions to problems in five core areas: agriculture, climate change, health care, learning disabilities, and governance. Teams with ideas at the conceptual stage received ₹5 lakh, while those with ready prototypes were awarded ₹25 lakh.
Efforts to build capacity
The ministry also launched the AI Competency Framework for public sector officials to build capacity among government personnel regarding the life-cycle of AI development and implementation.
In addition, an AI component was added to the government’s online course platform for civil servants, iGOT Karmayogi, which offers over 1,900 online courses. The new iGOT AI will use domestically developed AI models to personalise course recommendations for civil servants based on their ministry, role, and title, said Adil Zainulbhai, chairperson of the government’s Capacity Building Commission. The ultimate goal is to make iGOT easy to use so officials can simply enter questions, and the platform will suggest the most relevant courses.
“1.3 million civil servants have signed up for an AI course, and 940,000 of them have completed a course on AI,” Zainulbhai said. He added that the Commission’s goal is to ensure every civil servant in the country takes at least one course on AI and becomes “conversant in it” by the end of the year.
Vaishnaw also announced the call for applications for the IndiaAI Startups Global Acceleration program where, in collaboration with France’s Station F and HEC Paris, selected startups will spend four months at Station F’s accelerator programme and get access to European investors and markets.
Principal Scientific Advisor Ajay Sood said that AI faces three main challenges — skills development, computing capacity, and quality datasets — which were addressed through Thursday’s announcements.
“India has been on the right track on AI and data,” he said. Acknowledging privacy, safety, and ethical concerns related to AI, Sood said that the final report on AI governance would soon be released after reviewing all comments submitted during the consultation period. He added that India will contribute a “techno-legal framework” to the regulation of AI under the Global Data Framework mentioned in the UN report, similar to the measures recommended in India’s Digital Personal Data Protection Act.
