Hi, I'm Neelesh. I'm building an agentic financial stack — one balance that pays, settles and earns at the same time.

Money lives in separate systems that don't talk to each other — one to pay from, one to hold in, one to earn on — and every hop between them leaks a fee and a few days. Cross-border makes it obvious: an INR to AED payment still routes through dollars, so you pay a spread on the way in and another on the way out.

Krypton collapses those systems into one balance, so money keeps moving and keeps earning at the same time. Clark sits over it as the interface, for a person asking in words or an agent asking over MCP.

Cooking for myself through lockdown, I kept noticing that a dish is a short sequence of operations — boil, blend, sauté — in the right order at the right settings. Which means you don't want a robot chef. You want a compiler.

So the hardware stayed dumb and cheap: single-purpose modules, one job each. The intelligence went into the software, which turned a dish into a graph of steps and ran whatever could run at the same time. Fifteen people, two patents and paying subscribers, all of it alongside the Goldman job. We never raised a rupee.

I was an engineer on Transaction Banking's data and analytics platform. The systems underneath it were built to move payments, not to answer questions about them.

So most of the work was getting numbers out of them fast enough to be worth having, and clean enough that Liquidity, Risk, Finance and Sales would act on them without checking twice. Balances, FX, four payment rails, revenue and risk — six years of it, close enough to see how each piece behaves under load.

The medicine exists and the doctor exists. What fails is the road between them, and fixing roads does nothing in the ten minutes that decide the outcome. So we went over it — a drone a responder summons from an app, route planned before takeoff, no pilot in the loop.

The airframe turned out to be the easy part. The hard part was the clinical protocol: what to carry, in what, for how long. Which is why five doctors sat with the twenty engineers. Nobody was paid and everybody had classes.

India has the youngest workforce in the world, and most of the schools serving it never teach a child to make a computer do anything. That gap doesn't close on its own; it compounds for forty years.

An engineering campus has exactly one thing in surplus — undergraduates who can already code and have free afternoons. Writing the curriculum took a weekend. Getting eighty of them to the same school every week for three years took everything else.