Bharat AI exists because the gap between people who can use AI tools and people who understand how they work is widening, and the material that closes that gap is mostly locked behind subscriptions, university fees, or a wall of research papers.
A structured, free curriculum that takes someone from no background at all to genuinely understanding how modern AI systems work internally — not how to call an API, but what happens inside when you do.
Every lesson pairs a plain-language explanation with an interactive lab you can poke at. The belief underneath the whole project is simple: you do not understand a mechanism until you have broken it. So we let you break things.
Free to read, always. Every lesson currently published will remain free and open, with no account required. If Bharat AI ever adds paid offerings — cohort courses, mentoring, certification — they will be additions alongside the free material, never a wall placed in front of it.
Honest about difficulty. Some of this is hard. We will not pretend backpropagation is easy or that you can understand transformers in an afternoon. We will, however, never make it harder than it needs to be by hiding behind notation.
Corrections welcome. Technical writing accumulates errors. If you find one, please say so — corrections are published, not quietly buried.
The site is deliberately plain: static HTML, CSS and JavaScript, no build step, no framework, no tracking scripts. That means it loads fast on a slow connection, works on an old phone, and will still work in ten years.
The interactive labs are hand-written canvas and JavaScript. The Python playground runs CPython compiled to WebAssembly, entirely on your own machine — code you write in the playground never leaves your browser.
The project needs several kinds of help, and not all of them require writing code:
Corrections, questions, contributions, or a note that a lesson finally made something click — all welcome.
Email: hello@bharataiai.com
Site: www.bharataiai.com
Bharat AI is built in India and pays deliberate attention to problems that matter here — why Indian-language text costs more to process, why tokenizers trained on English corpora disadvantage other scripts, what India's DPDP Act requires of anyone deploying a model. The material itself is for anyone, anywhere.