Builder, Kalmantic Labs · In AI since computer vision
Bangalore · San Francisco
No AGI without peak tokens per watt per dollar.
Next chokepoint battle is open inference.
You know an answer is good when a child can repeat it back. That is the test I trust most.
In seeking better explanations, I keep asking why, down to first principles, through the mental models, until what is left is a metaphor a child can hold. My nephew Avnish keeps me honest. He will not pretend to follow.
These began as books for him. The quakeview ideas that moved the ground under me, Taleb, Naval, Housel, Fred Wilson, small enough to hand to a kid. I print them for friends. Online, they are free.

The first, made during COVID. The whole Incerto, risk and uncertainty, small enough for a child to carry.

Grew out of a post I wrote, hunting for anyone who thinks better than Naval.

An interpretation of Psychology of Money, not a summary. A gaze view of life against a goal-driven one.

Avnish asked what a VC does. Twenty years of Fred Wilson, but the 100 bets after taught me more.
Computer vision was my favourite subject at IIIT-H. Later I built a startup, hand-crafting SIFT vectors.
The work moved through Intuit, through enterprise software, through a few AI winters. Then the timing caught up, and now everyone can see it.
I am back to building. The lab is Kalmantic Labs.
I still make the bamboo shoot pre-seed bets that SV ignores, at Upekkha. Many companies in so far. The early end of the market is where I learned to see.
The writing here runs alongside the lab. Essays for the people building into this.
The Kalman filter is where I learned to find signal in noise. Kalmantic carries that lineage.
An Agentic Lab working on inference economics, agent infrastructure, NemoClaw strategy, and the long arc of systems that think.
Code. peakinfer and peakweights. Papers. SecLens with Subho at MatterSec Lab; PeakWeights with Vamshi Ambati.
Also Agentic Operator, Model is the Easy Part, and Agentic Enterprise. Five books so far, more at kalmantic.com.
Small experiments that escaped containment.
Also founderchart, ideaeval, and docopic.