MVP Engineering

His words, hedged the way he hedges them — "almost, around ₹1 crore" — on a single app, before COVID. An education platform. He went in wanting a thousand features. He should have tested one question: if I record a course, will anyone actually buy it from me? The feature list grew, the build dragged, he hired team after team, and — in his own words — a lot of people made a fool of him.

He has been frustrated by developers his whole life. That is the first sentence of the course.

Then he did a product management course at an IIM, to find out how a product is actually built. And what he teaches here is exactly that combination, stated in his own words: a self-learn mechanism, delivered through a systematic approach. The raw material is what he worked out by flailing. The delivery is what the course adds.

The first hour is not about AI at all. It is about paper.

Write every feature you want by hand. Don't write on the phone — you think harder on paper. Give it thirty minutes. Then open ChatGPT, switch on voice mode, and say: here is what I want to build, now tell me what else could go in it, and grill me. Go back to the paper and cross things out. Sort what survives into build, outsource, eliminate, delay.

Only then does the PRD get written — audience, problem, features, flows, pages. And one detail almost everybody gets wrong, which he states twice: your metrics do not live in the PRD. You decide them at the PRD stage and you keep them somewhere else.

Then pricing, which is where most non-technical founders are simply guessing. His rule is one sentence: if you do the same thing everyone else does, undercut them; if you genuinely differentiate, charge premium; and pricing the same as an established player means nobody has any reason to consider you. He works a lifetime deal all the way out — ten dollars a month is a hundred and twenty a year is three hundred and sixty over three — and gives you the retention number underneath it: hold the customer eight months and it is profit from there.

And then he builds two things on camera, without cutting.

Nearly three hours on a ChatGPT clone that runs locally on his own machine — including the part where he works out that the third feature is impossible, live, with arithmetic: about two minutes of compute for one good image, times twenty-four frames a second, is roughly an hour of his laptop for one second of video. So that feature dies, on screen, for a reason you can follow.

Then an hour and forty on an SEO blog site that ends up generating 673 articles in a single session — each with images, internal links, plus the privacy, terms and about pages and a working login.

The expensive lessons are all in there too. The API key his team leaked, which the platform auto-blocked only after ₹2.5 lakh had gone — "it could have been more, our credit limit was open." The developer quote of ₹10 lakh with the justification "the code will blow up, that's why", which you cannot argue with because you cannot audit it. The 193-star floor he uses before trusting anything off GitHub. And the comparison he is careful to state as his own experience: products he builds himself in four or five hours, his experienced developer took twenty days.

What this course does not do. Nothing here goes live. Both builds run locally; there is no deployment lesson, no hosting, no domain, no launch. And 83% of the runtime is two long, unedited sessions — no chapters, no summaries, no highlights reel. If you want polished, this is not polished. It is what actually happened, at the speed it happened.

Taught in Hindi.