I had what felt like a god-tier idea.
I would keep pouring every mistake, decision, person, system, and changed opinion into this website. Then I would connect the whole thing to an llm. A client, employee, or founder could ask, “what would Moksh do if he were me?” and get a mini version of me without waiting for me to finish another coffee.
There is one small issue. I tried to build the person before I had written enough of the person down.
The technical part works. The mcp lets an AI search the website, open a note, and cite the page it used. That is useful. It is a searchable notebook.
But searching my sentences and making a decision like me are two completely different jobs.
We tested it with real questions. What should I do with an underperforming co-founder? Should I hire more people to move faster? How should I compare risk? What should I do when the product works but no customers are coming?
It confidently fetched notes about pricing, break days, and agency finances. Technically, it answered. Practically, it had walked into the wrong room and started giving a presentation.
The live website has 21 notebook notes. This review version has 49, and many are short or still waiting for my approval. There are opinions in here. There are not enough proper decision records with the context, constraint, action, outcome, confidence, and date.
That means mini-Moksh can tell you what I have said. It cannot reliably tell you what I would do in your exact situation. Claiming otherwise would be me putting a lab coat on a cardboard cut-out.
So am I killing it?
No. I am moving it out of the centre of the homepage and putting it where it belongs: the lab.
The website will first become a public operating log. Whenever I make a serious decision, I want to record the problem, what I knew, what I chose, what happened, and whether I still believe the same thing. Mini-Moksh can earn its way back to the centre when the source material becomes deep enough and the answers survive a blind test.
Until then, the honest label is: ask my public notebook. Early experiment.
This might be failure number whatever
I could spend months documenting everything and still discover that no one wants a mini version of me. Honestly, fair. I am twenty. Maybe the world does not urgently require another twenty-year-old giving it advice.
But if the experiment fails, the writing still remains. My team gets clearer decisions. I can see when I changed my mind. Other founders can steal one useful mechanism. The failure leaves behind an asset.
That is the kind of thing I want to fail at.
Build the smallest version. Ask reality whether it is useful. If reality says no, do not make the button bigger. Change the idea.