ChatGPT for education? IIT Madras director on how Bodhan AI will work and what it can do

In the lead up to the AI Impact Summit 2026, the ministry of education and Indian Institute of Technology (IIT) Madras launched Bodhan AI, late last week. The “sovereign AI stack” – fully homegrown, from data to tech – is intended to be a digital public infrastructure (DPI) which, when ready, “is intended to be as effective as foundational models like ChatGPT”, said V Kamakoti, director, IIT Madras whose Centre of Excellence (CoE) in Artificial Intelligence (AI) for Education is tasked with building it.

At its launch, education minister Dharmendra Pradhan used the example of the National Payments Corporation of India (NPCI) to explain the role Bharat Bodhan AI will play. The NPCI’s Unified Payment Interface (UPI) is the DPI enabling payments and transfers through mobile phone applications. Bodhan AI’s infrastructure will be ready in six months, said Kamakoti and while no formal agreements are in place yet, the plan is to have four states use it at first.

Bodhan AI was launched as a Section 8 non-profit company. “The objective of this company is to basically ensure equitable, accessible, affordable, quality education for all. This is our sustainable development goal,” said Kamakoti.

As many as 120 education technology (edtech) startups exhibited at the two-day Bharat Bodhan AI Conclave held last week.

What is India's Education Sovereign AI Stack?

The NPCI and UPI provide a useful analogy to explain Bharat EduAI Stack or India’s sovereign AI stack.

Kamakoti stated, “NPCI built the fundamental infrastructure for digital payments in India. However, the general public doesn’t use NPCI directly but uses apps like PhonePe or Google Pay. But underneath all of them sits UPI, the shared infrastructure that made digital finance accessible to everyone with a smartphone.”

Similarly, IIT Madras’ Bodhan AI – a company and equivalent to NPCI in this analogy – is Bharat EduAI Stack, a digital public infrastructure for learning.

“Instead of financial inclusion through UPI, the goal here is educational inclusion through a sovereign AI Stack DPI hosted in India,” Kamakoti added.

Students won’t log into Bodhan AI or the Bharat EduAI Stack directly but through apps and tools built by schools, states, and edtech startups. If it goes to plan, these tools could analyse student responses, give personalised feedback, flag learning gaps early, support teachers with classroom insights, and even help with mentoring or career guidance.

AI in Education: Where do edtech startups and state apps come in?

First, a basic foundation code will be created by Bodhan AI, Kamakoti explained, “Just like UPI functions as shared infrastructure. This foundation will be the Sovereign India AI Stack.”

On top of this stack, startup companies will build applications. “These startups will use the infrastructure to launch highly-scalable software and tools aimed at enabling quality education for the poorest of the poor,” stated Kamakoti. These tools and apps – AI-in-education’s equivalent of PhonePe and Google Pay – will be the end users way in and the most familiar part for students, teachers and other stakeholders.

How does Bodhan AI and the Sovereign India AI Stack connect with teachers and students?

“When you buy bananas, the shopkeeper shows you a Quick Response Code (QR) code,” explained Kamakoti. “You open your Google Pay, scan it, and the transaction happens in seconds. You don’t see the backend but UPI processes the payment, moves money from your account to the seller’s account, and everything works smoothly. Apply that to education. An XYZ edtech startup will create an app. Students, parents or teachers will download that app on their phones. Suppose the app is built to understand how well a student reads aloud. A teacher clicks “record,” and the student’s voice is captured.”

Once the voice is recorded, the app sends it for processing. The software may analyse pronunciation, fluency, pauses and then give feedback. But for this to happen accurately and quickly, a lot is required behind the scenes. “Digital infrastructure like compute power, cloud infrastructure, central processing units, large-scale processing capacity and others [inputs] are required,” Kamakoti said. Further, the voice must also be processed in the student’s mother tongue.

That is where the Bodhan AI comes in. Its EduAI Stack will power the heavy processing after a student records their voice. It provides the core AI infrastructure needed to process that audio, generate insights, and send the output back to the user.

In place of a QR code, the edutech startup will employ voice tools.

Will Bodhan AI have its own large language model (LLM)?

“Bodhan AI will provide a very large language model (LLM) that is intended to be as effective as foundational models like ChatGPT,” said Kamakoti. He explained that the Bodhan AI’s LLM will function as a basic foundational model that others can further customise.

The model will be accessible to the public, said Kamakoti. “The intention is to offer it as digital public infrastructure — similar in spirit to systems like UPI and Aadhaar — so that it is not restricted to a closed group of institutions,” he said.

What kind of data will Bodhan AI analyse?

Kamakoti said that the Bodhan AI is intended to analyse multi-format data. This includes data from optical character recognition (OCR), attendance records, and other structured and unstructured formats.

In nutshell, the system is designed to process different types of educational data — not just text, but multiple formats relevant to learning and classroom environments.

How will this work in places with few devices or limited connectivity?

To address accessibility in remotest parts of the country, Bodhan AI is now working on Frugal AI. “Frugal AI refers to AI systems that are designed to operate with less money, less hardware and limited infrastructure, while still delivering meaningful outcomes. This is the real game of Bodhan AI,” added Kamakoti. “The aim is to work with limited resources because we have to scale to a 1.5 million population.”

How will the startups be onboarded?

Onboarding will depend on need. “If a startup wants to interact and engage with Bodhan AI, and if there is alignment with the ministry of education’s priorities, then onboarding can take place. The key question is whether the startup actually needs the infrastructure. If a company already has a solution that is scaling effectively on its own, there may be no reason to integrate,” Kamakoti said.

However, if a startup requires support, then Bodhan AI can assist them – with the larger objective of getting something in return in the advancement of equitable, accessible, affordable and quality education, he added.

Will these startups build on a common API layer or will they integrate into a central marketplace?

If startups choose to associate with Bodhan AI, they will be provided access to the DPI, Kamakoti stated. “This infrastructure will include a set of Application Programming Interfaces (APIs) that organisations can use to build their applications. In that sense, the model is based on a shared API layer within the digital public infrastructure of Bodhan AI, which participating startups can integrate into their systems.”

Kamakoti added that during the two-day Bodhan AI conclave, around half a dozen states participated, including Maharashtra, Andhra Pradesh, Odisha, Karnataka and Goa. These states are expected to be the end users.

“Once the digital public infrastructure (DPI) is in place, the plan is to work closely with these states so that their existing apps and tools can use the platform’s APIs. Through this integration, states can launch multiple applications. The effectiveness of these tools will then be assessed through qualitative and quantitative feedback,” he explained.

At this stage, however, there are no formal commitments but the ministry of education has received strong interests from participating states, Kamakoti said.

How will you ensure that this specific DPI will be used only for educational purposes?

Safeguards will be built into the basic infrastructure itself. “At the time of onboarding, there will be clearly defined use cases that participating organisations must adhere to. The expectation is that the infrastructure will be used strictly for educational purposes, and this can be ensured through onboarding conditions,” said Kamakoti adding, “there will also be audit mechanisms to monitor how the platform is being used. This is intended to prevent misuse or diversion of resources for non-educational purposes.”

Kamakoti explained that the approach is comparable to how the app stores review and analyse applications before making them available. Similarly, applications using this infrastructure will go through basic checks.

The monitoring and governance systems are expected to evolve as the platform grows, he added.

What is the timeline for the implementation of Bodhan AI?

The aim is to have the DPI infrastructure ready within the next six months. “The plan is to make it available and work with at least four states initially. These states would use the infrastructure, develop their own applications on top of it, and deploy them within their systems,” said Kamakoti.

The Bodhan AI Conclave was designed as an “exploratory workshop”. “It focused on understanding who needs what, who has what, and how those needs and capabilities can be aligned. The objective was to identify how different stakeholders, including states, institutions and startups, could come together around and work together on a shared infrastructure,” stated Kamakoti.

There is confidence that many states, and eventually almost every state, will adopt it, just like UPI has become widely used over time.