Indian students rank among the top students almost anywhere in the world. The joke is that, from the womb to the afterlife, they are taught to study maths and science which serves them well throughout life. A report by Redseer says that India’s test-preparation market has been estimated at $14.8 billion in FY26, projected to reach $23-$26 billion by FY30 (12%-15% CAGR), with around 2,00,000 coaching centres nationwide. There are 65 million students in classes 9-12, of whom around 27 million attend government schools. A 2025 National Statistical Office survey found that 27% of students were taking or had taken private coaching in the academic year (30.7% urban, 25.5% rural). Across datasets, coaching incidence is around 27%-30%, implying that 17 million to 20 million students in classes 9-12 participate in some form of test-prep coaching.

No single private platform has the capacity or curriculum breadth to serve as the sole national provider. PhysicsWallah, for example, has around 4.9 million paid online users — while the target population is 10 to 100 times larger. This needs to be built as a platform and network play, not a vendor contract.

Better learning, better outcomes

How can the government take the help of Artificial Intelligence (AI) to deliver a free coach and tutor to every Indian child? The objective should not be to bring down the private coaching industry but to improve the learning standards and examination outcomes of the last child in the line who today spends thousands of rupees on dubious coaching centres without achieving the necessary results. Bring the industry in as suppliers, or they will lobby against it and the reputational risk of failing completely, is high for the government to sustain.

We should not nationalise coaching. Build a public rail, private engines model: the state supplies the digital public infrastructure (identity, payments, discovery, data, credentialing) and aggregates demand; the private sector — star teachers, ed-tech firms, local tutors — supplies content and last-mile delivery shared by both. Government becomes the platform and market-maker, not the content producer. This is exactly how UPI beat closed payment apps and how the pen Network for Digital Commerce is attempting to beat closed commerce: unbundle the monopoly, open the protocol, let a thousand providers compete on an open network.

The Indian coaching industry charges high fees because it bundles four things that do not have to be bundled: content; doubt-solving/mentoring; peer group and discipline, a credible signal of quality. DPI lets us unbundle these and drive the marginal cost of the first two toward zero. This can be anchored in an existing centre of excellence such as Bodhan.ai at IIT Madras (which has a ₹500 crore, five year government grant) or a similar National Payments Corporation of India (NPCI)-like body that needs to be created for education (a Section 8 not-for-profit steward).

Where it should begin

It should start with the National Eligibility cum Entrance Test (Undergraduate), or NEET, and the Joint Entrance Examination, the two most widely taken examinations in India. JEE/NEET are objective, machine-gradable, and high-volume. The content is already freely available on YouTube; what is scarce is a trusted, standardised rank signal and personalised doubt-solving.

An IIT professor who teaches students who have cleared JEE 30 years ago may not be the best option for teaching test prep. Today’s test prep skill and requirement is in a very different age, time, and pedagogy. Several, including this writer, who have cleared the IAS, GMAT, GRE, and IIT examinations, would shudder to take it again, much less teach it.

Government aggregates demand thereby drives down the customer acquisition cost for the private sector and owns the rails and the trust layer; the private sector and star teachers compete to deliver content and mentoring on an open network; the student owns the data and carries a voucher (optional). That is how you turn a $15b paywall industry into free, plural, high-quality prep — without the state trying to become the teacher.

As with any private good using public rails, it will use Aadhaar/DigiLocker for student sign-up and to de-duplicate; APAAR ID to tie learning to the academic record. Along with this will be a curated content registry, an open catalogue where any accredited provider publishes lessons, problem sets, and mock tests against a standard syllabus taxonomy (NEET/JEE topic tree) and open metadata (language, difficulty, medium, accessibility tags). The government will set the taxonomy and standards for the content and ensure interoperability.

Content interoperability is the whole game — a student’s progress in one provider’s module must be portable to another’s. There shall be a discovery layer, a government-run reference app plus open APIs so third parties (including the DIKSHA app, state apps, even Google/YouTube) can surface the same content — multiple front ends, one network. For example, PhysicsWallah, Vedantu, Allen or Narayana could offer their material for free and charge for doubt-solving or premium sessions with ranked professors. IIT professors could also upload their lectures and be rated and ranked on the platform. Adaptive personalisation should be a core feature and as a network service, an open recommendation/diagnostic engine that routes say a weak-in-organic-chemistry student to the best-rated module on that topic, regardless of provider. These will be accompanied by a learning wallet and a progress ledger, a student-owned record of what they have studied, mock scores, and mastery per topic, consent-gated through the DEPA/Account Aggregator pattern. The student — not the ed-tech firm — owns the data, which is the single biggest structural break from today’s walled-garden apps. Lastly, while the platform will be mostly free at the point of use, UPI-plumbed premium add-ons (live mentoring, graded assignments) can be micro-charged or voucher-funded, should the government wish to create a DBT-like e-Shiksha wallet. Deliberately thin the government software: The state builds protocols and a reference app; it does not try to build the best learning app. It will not win that race and should not try this. Government alone can aggregate the demand and credential the supply and outcomes — that is all it should do. It should auto-invite every registered Class 11-12 student through schools, CBSE/State boards, and the DIKSHA install base. The system should guarantee reach through free data-light streaming, downloadable content for low-bandwidth areas, and distribution through the five lakh-plus Common Service Centres and school computer labs for students without personal devices. Equity is the mandate — the ₹60,000 crore industry already serves the affluent; the gap is everyone else.

Government should publish open ratings and outcome data: publish, per provider and per module, how students who used it scored on official mocks and the real exam. Quality signal becomes objective, not marketing. This is the credential the coaching industry currently monopolises. Government alone can ensure independent accreditation of providers (content accuracy, no misleading claims), with delisting teeth. Further there should be a no-lock-in rule to ensure that providers cannot hold a student’s progress data hostage; portability is enforced at the protocol level.

Models on offer

There is no dearth of content, good and bad, on YouTube but where an AI tutor helps is in practice, doubt solving and adaptive personalisation, which only AI can do. There are good open-source models such as Gemma, which give a million tokens for about 20 to 30 cents or 10,000 tokens per day or 3.6 million tokens per year for just ₹100. Further, every AI minute is equivalent to 1,000 voice minutes of a human; frontier tech companies such as Meta and Google are paying star Kota teachers $25 to $30 per hour for commoditising their content. Further, their models will get trained on it and Generative Inference will do the rest on top of it.

We should not be delulu — as Gen Z would put it — government should not become the AI tutor itself. It should only enable the AI tutor through the blueprint given.

Srivatsa Krishna is an IAS officer. The views expressed are personal