I can build a product with AI —and it actually works.
A year in AI: prompt, context, loop and graph engineering, automation and agent building. I don't write the code — I direct it, and I check what comes back.
“I don't write the code. I decide what gets built, hand the work to AI, and check what comes back — because the last call has to stay human.”
Four things I do, explained the way I would explain them to a friend.
Each card opens a page with the full idea, a real example from my own work, and a prompt you can copy and use today. No jargon, because the jargon never helped anyone.
Telling the AI exactly what you want
A role, the situation, the problem, and what you need back — the same four things you would give a person before asking them to act.
Open the library → 02 · ContextShowing before telling
A designer does not guess your style — you show the thumbnails you like and the ones that earned clicks. Context is handing that over before the work starts.
Open the library → 03 · LoopDoing the same thing until the job is done
Ten boxes, one trip each, and you keep going until the last one is across. A loop stops on a condition you wrote down, not when the AI feels finished.
Open the library → 04 · GraphMore than one pair of hands
A writer and an examiner, with an order between them — and an examiner the writer cannot instruct. Break the job up, and keep the final call human.
Open the library →Three things I built, and what each one taught me.
Every number below is a real count from the repository, not a business metric I cannot defend. Each one opens its own page.
Employee Hub — AI employees for any business
Five new AI employees every morning at 06:00 IST, each one a skill with a written standard for how it behaves when it cannot do the job.
TruePick — an affiliate storefront with an admin
A JSON-driven shop with a blog that reads itself aloud, YouTube reviews on product pages, and a WordPress-style admin panel behind it.
Social Media Manager — an agent skill
Finds which social connectors actually work, keeps a knowledge base of your brand voice, then plans, writes and schedules. MIT licensed.
One email when something new is built.
No weekly newsletter, no drip sequence, no tracking pixels. When a project ships or a discipline page gets a new example, you get one short email. That is the whole promise.
I'm not taking work.
Ask me anything anyway.
Everything here I built for myself, to learn what AI can and cannot be trusted with. If you want to know how one of these was made, ask — I will tell you the honest version, including what broke. Email is the fastest way to reach me.