Classes1–5
AI foundations
- Language and storytelling
- Visual exploration
- Basic problem solving
- Patterns and algorithms
- Humans compared with machines
- Introductory AI concepts
- Responsible technology habits
AI for India’s government schools · Classes 1 to 10
Every child learns at their own pace. Every teacher stays in charge.
CMDLearn is an AI learning platform we are building for government schools in India. Each student gets a tutor that works in their language, inside limits their teacher sets. Teachers get tools that show who needs help. Departments get proof of what changed, measured against where children started.
Live demo: a short Class 8 science class (forces) with Asha, a talking AI tutor, then a five-minute video call where you ask her anything. It is a private demo for parents, teachers and partners: you need an access code, so ask us for one. Your camera stays off unless you turn it on, and your picture never leaves your device.
Early stage · a demo, not a product yet We are still building. No schools use CMDLearn yet and we have not measured any results. This page explains what we are making, how we will test it, and how you will know whether it worked.
- What is AI?
- How does it work?
- How can it be wrong?
- How do I verify it?
- How do I use it responsibly?
- How do I build with it?
01 · The problem
The problem we are solving
One classroom. Children at ten different levels. One teacher. And nobody above the school can see whether anyone is actually learning.
Walk into a government school classroom and you will find children reading and doing maths at very different levels. One teacher handles all of them, plus the syllabus, the exams and the paperwork. Teaching each child at their own level is not a question of caring. It is a question of time, and of knowing where each child actually is.
AI is already reaching students, whether schools are ready or not. Left alone, it shows up as an open chatbot: no age limits, no link to the syllabus, no teacher watching, no record of what it told a child. That is a risk to the child and a headache for the department.
Most education software tells you how much it was used. That is the easy question. The hard one is: what changed, for which children, compared with where they started, and would it work again somewhere else?
- Mixed levels Children in one class can be years apart. Teaching to the middle leaves both ends behind.
- Slow feedback By the time exam results arrive, it is too late to change what happens next week.
- AI taught in bits AI comes as a one-off workshop, not something built up year by year from Class 1 to Class 10.
- Language Children learn fastest in the language they think in. Most software assumes English.
- Internet Anything that needs steady internet and a personal device leaves out the schools that need it most.
02 · Evidence
What the research says
We have read the policy and the evidence, including the parts that argue against us. Here is what it says, with links. These are other people’s findings, not endorsements of CMDLearn.
What Indian policy asks for
-
The National Education Policy 2020 supports technology in education “provided these interventions are rigorously and transparently evaluated in relevant contexts before they are scaled up”, and calls for “carefully designed and appropriately scaled pilot studies”.
National Education Policy 2020, paras 23.3 and 24.1 · Government of India Read the policy
-
The National Curriculum Framework for School Education 2023 makes “Mathematics and Computational Thinking” a named curricular area, and counts digital literacy, coding and computational thinking among the capacities all students should acquire.
NCF-SE 2023 · NCERT, Ministry of Education Read the framework
-
CBSE has introduced a Computational Thinking and Artificial Intelligence curriculum for Classes III to VIII, designed “without dependence on any specific platform” and taught by existing subject teachers rather than new staff.
Circular Acad-15/2026 and the CT&AI framework · CBSE Read the circular
What works, and what does not
-
The Global Education Evidence Advisory Panel rates teaching targeted to a child’s learning level, rather than their grade, a “Great Buy”. It places personalised, adaptive software a tier lower, as “promising but limited evidence”, with “very limited evidence at scale or with governments implementing”. Buying hardware alone is listed among its “Bad Buys”.
Smart Buys, 2023 · GEEAP (FCDO, UNICEF, USAID, World Bank) Read the report
-
Across randomised trials in low- and middle-income countries, software that adapts to a learner’s level showed a pooled effect of 0.35 standard deviations, against 0.13 for software personalising only interests or feedback. The authors caution that all the trials studied supplementary instruction, and that whether personalised technology can be scaled cost-effectively “is not yet known”.
Major, Francis & Tsapali, 2021 · British Journal of Educational Technology Read the meta-analysis
-
In Indian government schools, giving teachers the materials for level-based teaching, or the materials plus training, produced no significant effect on its own. What worked added a mandated dedicated hour, physical regrouping by level, and coordinators who mentored and monitored teachers on site.
Banerjee et al., 2017 · Journal of Economic Perspectives; J-PAL case study Read the case study
-
ASER’s 2024 visits to 15,728 government schools found that two-thirds of Std I and Std II classrooms were multigrade, with children from more than one grade taught together. The spread of levels inside a single room is a measured condition, not an assumption.
ASER 2024 national findings · ASER Centre / Pratham Read the findings
What is still unproven about AI in classrooms
-
UNESCO presents every educational use of generative AI in its 2023 guidance under the heading “potential but unproven uses”, and writes that “it is unlikely that GenAI alone will solve any of the problems facing education systems around the world”.
Guidance for generative AI in education and research, 2023 · UNESCO Read the guidance
-
The same guidance sets an age limit of 13 for the use of AI tools in the classroom and calls for a human-centred approach to AI in education. Our advice, not UNESCO’s: an adult should check what these tools tell a child.
Guidance for generative AI in education and research, 2023 · UNESCO Read the guidance
-
UNESCO also notes that widely used models are trained on data reflecting the norms of the Global North, making them “inappropriate for locally relevant AI algorithms in data-poor communities”, and asks that training data include local and indigenous languages.
Guidance for generative AI in education and research, 2023 · UNESCO Read the guidance
Three of these findings argue against products like ours. They are here because you would find them anyway, and because they set the bar: a tutor with limits, not an open chatbot. Support for teachers, not software dropped on a desk. Indian languages built in, not bolted on. And proof, not promises.
03 · Platform
What CMDLearn does
One platform, four views. The student, the teacher, the school and the department each see what they need from the same learning record.
-
AI tutor
A tutor that explains, asks questions and checks understanding, inside the topics and age limits the teacher has set. It does not just hand over answers.
-
Teacher assistant
Lesson plans, practice sheets, worksheets and feedback drafts, ready for the teacher to use, edit or throw out.
-
Personalised learning
Each child works on what they are ready for, not on what the class average has covered.
-
Tests and progress tracking
Tests at the start, along the way and at the end, mapped to skills. Progress reads as what a child can now do, not just marks.
-
Projects and simulations
Hands-on work where students build things, test them and see where they break.
-
AI literacy, safety and ethics
Age-appropriate lessons on how AI works, how it goes wrong and how to use it responsibly, from Class 1 to Class 10.
-
Teacher dashboard
See which children are stuck, who needs attention first, and what has already been tried.
-
Department dashboard
School, block, district and state views, reported against the measures you agreed on, with access by role.
-
Indian languages and voice
Content and tutoring in Indian languages, starting with Marathi, Hindi and English. Voice first, so early readers can talk instead of type.
-
Built for weak internet
Built for slow and patchy internet. Offline lessons and tests that sync when a connection comes back are planned for a later release.
04 · Curriculum
One journey, Class 1 to Class 10
From first ideas about AI to using it as a co-pilot, with the same six questions asked at greater depth every year.
Classes6–8
Applied AI
- Data and algorithms
- Machine-learning concepts
- Generative AI
- Coding and research
- Simulations
- Critical thinking
- AI ethics and safety
Classes9–10
AI as co-pilot
- Python and data analysis
- AI and machine-learning fundamentals
- Generative AI and prompting
- APIs and automation
- Scientific reasoning
- Building AI applications
- What is AI?
- How does it work?
- How can it be wrong?
- How do I verify it?
- How do I use it responsibly?
- How do I build with it?
The same six questions come back every year, answered at greater depth each time. The third and fourth matter most. A child who cannot explain how AI gets things wrong, or check an answer against a source, is not AI-literate, however good their prompts are.
05 · Learning loop
How the learning loop works
Six steps. The teacher is step four, and every decision about a child is made there.
Student
A student works on a lesson, practice set, test or project made for their age.
AI tutor
The tutor helps, within the topics and safety limits set for that class.
Learning data
What the child can and cannot do yet is recorded against the skills map.
Teacher
The teacher sees the gaps, the progress and suggested next steps, sorted by who needs help first.
Action
The teacher decides what happens next. The child gets targeted support.
Department view
Summed-up, role-based views help administrators decide whether a programme should continue, change or grow.
Nothing in this loop skips the teacher. The platform suggests. The teacher decides.
06 · Pilots
How a pilot works
A pilot is a measurement exercise with a platform attached, not a rollout with a survey at the end.
Measure first
We measure where children are before anything is installed, using a method the department has agreed to in writing.
Train teachers
Teachers learn the platform, the safety rules, the dashboards and how to act on what they see.
Use in class
Students use CMDLearn alongside normal teaching. It fits the timetable; it does not replace it.
Check in
We review the agreed measures at fixed points and fix problems in the open.
Measure again
We run the same tests as at the start, so the comparison is like for like.
Report
What changed, what did not, what went wrong, and whether the evidence justifies going bigger.
We are proposing pilots with state and district school systems across India. Measuring first is not a formality. Without a starting point there is nothing to compare against, and no change can honestly be called a result.
What we ask of a department
- One officer who can make decisions for the pilot
- Sign-off on which schools, classes and teachers take part
- Agreement on how we will measure, before anything is installed
- Clear rules on data and privacy, in writing
- Time for teachers to be trained
- A clear picture of the devices and internet in the chosen schools
- A calendar for check-ins. Nothing starts until these are settled.
What we will measure
We agree the measures before launch. There are no numbers here on purpose: a target set before the first measurement is decoration.
- Learning How much children’s skills improve from where they started.
- AI literacy Whether children understand how AI works and how to use it safely.
- Teacher use Whether teachers actually use it, and act on what it shows them.
- Teacher time Time saved on preparation and marking.
- Participation Lessons finished, practice done, projects completed.
- Fairness Take-up across girls and boys, schools and locations.
- Reliability Uptime, offline sync, and how fast we fix problems.
- Value to the department Reporting time saved, and how clearly you can see learning gaps.
Targets are set after the first measurement, never before. The report is written to be used against us: it will say what failed, where uptake was thin, and which claims the data does not support.
We will not claim results we have not measured. A pilot that shows no difference is a real finding, and we would publish it as one.
07 · Safety
Keeping children safe
The users are children. Safety shapes how the product is built, not a policy page stuck on afterwards.
Teacher in control
Teachers choose what their class can access, can see what the tutor was asked, and can override any suggestion.
Built for each age
What a Class 2 child can do is deliberately narrower than what a Class 9 student can do.
Guardrails
The tutor sticks to the syllabus and safe topics. Unsafe or off-topic requests are refused and passed to an adult.
Healthy scepticism
Children learn that AI can be confidently wrong, and are asked to check answers. Safety is learned, not just enforced.
A human in the loop
Anything worrying goes to the teacher and school coordinator, with a clear route if a child may be at risk.
Minimal data
We collect only what the agreed purpose needs. Every field has a stated reason and a deletion date.
Access by role
Each role sees only what it needs. Department views never show individual children.
A full record
Access, admin actions and any change to the AI are logged, so a reviewer can reconstruct what happened.
Incident plan
A written playbook for safety and data incidents. The department is told, and a record is kept.
Child data, by the book
Designed around India’s data protection law and departmental rules, with legal review before any real student data is collected.
08 · Roadmap
The two-year plan
Four phases, in months. Each one has to produce something the next can be judged on.
Months0–6
Design
Curriculum, skills map, tests, technical design, first-language content and the safety model, all agreed before code is built around them.
Months6–12
Build
Student app, teacher and department dashboards, testing engine and AI engine. Working offline is a requirement, not a feature.
Months12–18
Pilot
A small number of schools. Measure first, train teachers, then revise the product and curriculum against what classrooms actually do with them.
Months18–24
Expand and prove
Widen the pilot, measure again, publish a report with its method, and prepare what a department needs to decide on a bigger rollout.
This is a plan, not a record. We will publish progress against these phases as they finish, including the ones that slip.
09 · About
What CMDLearn is, and is not
Learning infrastructure for India’s government schools, built on one rule: AI supports the teacher, and the teacher decides.
CMDLearn is a startup building AI learning tools for government school systems across India. We design for the full range of languages, internet conditions and classroom realities found across the country, not for one region.
That one rule shapes everything. Every suggestion goes to an adult who knows the child. Nothing reaches a student outside the limits their teacher has set. What a school system gets is a product plus a way to roll it out, with training and measurement included. Not a staffing contract.
There are no results, logos or testimonials on this site because we do not have any yet. When we do, we will publish them along with how we got them.
CMDLearn is
- Curriculum, software, tests, teacher tools and a department dashboard
- Small, measured pilots with a start test and an end test
- Results compared with where children started, published in full
- Something a school can run with the teachers it already has
- Multilingual, voice first, and works on bad internet
CMDLearn is not
- A coaching-class or tuition business
- Daily classes run by our staff
- An open chatbot handed to children
- A usage chart dressed up as a result
- A claim we have not measured
10 · Questions
Questions people ask us
The things a department, a partner or a reviewer usually wants to know first.
Has CMDLearn run a pilot or been used in schools yet?
No. CMDLearn is still being built. There are no pilot results, no schools using it and no users to report. We are proposing pilots with state and district school systems, but none has run yet. If that changes, this page will say so, and we will publish how we measured it alongside the result.
Why are there no numbers or results on this website?
Because we have not measured any yet. A lot of education technology publishes made-up example figures as if they were findings. That makes a company harder to trust, not easier. Numbers will appear here only when they come from a real start-and-end measurement that a department has agreed to.
What is a pilot, exactly?
A fixed sequence. Measure where children are, train the teachers, use the platform alongside normal lessons, check progress at agreed points, measure again with the same tests, then publish a report. The comparison is always against the starting point the department approved before anything was installed. Without that, no claim of improvement can be checked.
How would a department know if a pilot failed?
The end test uses the same method as the start test, so a flat or negative result stays visible. The report will say where use was thin, which parts went unused and which claims the data does not support. A pilot that shows no difference is a real finding, and we would report it as one.
Can children chat freely with the AI tutor?
No. The tutor stays inside the syllabus and the age level and topics the teacher has switched on for that class. Unsafe, upsetting or off-topic requests are refused and passed to an adult. That limit is the difference between an education tool and an open chatbot handed to a child.
Does CMDLearn replace teachers?
No, and it is built so it cannot. The platform produces signals, drafts and suggestions. The teacher accepts, edits or ignores them and decides what happens in class. What it replaces is the manual work of finding learning gaps and preparing different material for different children.
Which languages does CMDLearn support?
Indian languages, starting with Marathi, Hindi and English, with more added as content and tests are built. Language is a core part of the product, not a translation layer, because children learn fastest in the language they think in. Voice comes first, so early readers and children sharing a device can speak instead of type.
What happens in schools with poor internet, or none at all?
The platform is built for weak connections first. The first version works on slow connections, and offline use with sync when the internet comes back is planned for a later release. How well it works is one of the things we measure in a pilot. Reliability in low-connectivity schools is part of the result, not a footnote.
What student data would CMDLearn collect, and who can see it?
Only what is needed for the agreed educational and monitoring purpose. Every field has a stated reason and a deletion date, and nothing is collected until data rules are agreed in writing with the department. Access depends on role: teachers see their students, school staff see their school, administrators see summaries. Individual children never appear in department-level views.
What does CMDLearn ask a department to provide?
One officer who can make decisions, sign-off on the schools and teachers taking part, agreement on how we will measure before anything is installed, written rules on data and privacy, time for teacher training, a clear picture of the devices and internet in the chosen schools, and a calendar for check-ins. Nothing starts until those are settled.
Is CMDLearn a coaching class or tuition product?
No. CMDLearn is learning infrastructure: curriculum, software, tests, teacher tools, dashboards, training and measurement. It is not daily classes run by our staff. It is deliberately light on field staff, so a school system can run it with the teachers it already has.
Is CMDLearn approved, funded or selected by any government body?
No. CMDLearn makes no claim of approval, funding, selection or purchase by any government body, and nothing on this site should be read that way. Any future work with a department would go through that department’s normal evaluation and procurement process.
Founding prices
Plans for families and schools
Plans open at launch. No payment is taken yet: every button below joins the waitlist with the plan you picked, and nobody pays until early access opens and you choose to.
-
Free start
Any class, 1 to 10
₹0
Will be free every day, with a daily limit
When it launches, try the AI tutor in Hindi, Marathi or English. It will guide a child to the answer instead of handing it over.
Join the waitlist, free -
Junior
Class 1 to 5
₹199 a month
or ₹1,499 a year (save 37%)
When it launches: voice first, short sessions and games. A parent or teacher will start every session and see what happened afterwards.
Choose Junior -
Senior
Class 6 to 10
₹299 a month
or ₹2,499 a year (save 30%)
When it launches: NCERT chapters as spoken lessons with a game at each step, and a short report for the parent.
Choose Senior -
Schools
Private schools, from 100 students
₹600 per student a year
Founding offer: first term free, then ₹420 per student for year one
When it launches: for classrooms and computer labs, with a teacher view. Prices exclude GST. Larger schools and NGOs, talk to us.
Join the waitlist
A second child on a family plan adds ₹99 a month. Prices are in Indian rupees and are founding prices for early families and schools: we confirm the price with you before anyone pays.
We sell a learning tool, not classes. Meanwhile, read our free guides or print the free worksheets for Class 1 to 10, in English, Hindi and Marathi, with answers.
Guides parents and teachers ask for most: is an AI tutor safe for my child?, ChatGPT for homework without cheating, how to judge an AI study app, the CBSE AI curriculum for Classes 3 to 8 and AI in the Indian classroom, for teachers.
11 · Contact
Talk to us
Write to us directly. Hard questions are welcome, including the awkward ones.
If you are looking at CMDLearn for a school system, tell us which classes, which languages your students learn in, and what the internet is like in your schools. If there is something specific you would want measured, say so early. It is much easier to agree before a pilot than to argue about after one.
-
General questions and due diligence
[email protected]Questions about the product, requests for documents, and anything you need to check what this page says.
-
State and district school systems
[email protected]Scoping a pilot, how we would measure it, data rules, teacher training and the check-in calendar.
-
CSR and implementation partners
[email protected]Funding a rollout in government schools, or delivering one alongside a department.
Parents and schools: free to join. We will email you when early access opens.
Join the waitlist