A new company, self-funded, built by one person. We would rather you knew that from us than worked it out later.
KritiGrid is an AI and analytics startup working on one problem: pricing the curtailment risk a large electrical load takes on when it accepts flexible interconnection. It was founded in 2026 by Sirisha Gottipati and is funded entirely out of pocket. There is no institutional backing, no team of engineers, and no customer list to show you yet.
We are telling you this on the page where most companies would imply otherwise, because the whole argument of this product is that you should check our claims rather than trust them. That argument does not survive an inflated About page.
A small company is a real risk and we will not argue it away. What we can say is that nothing here depends on us being trusted, or on us still being here. Every input is public. The analysis is deterministic and seeded, so the same command produces the same numbers on someone else's machine. The failure modes are published beside the results rather than discovered later.
The level drifts when the network changes. Our forecaster gets the hourly and seasonal shape right and the absolute level wrong in years when lines are reconfigured. In the two years we hold, 132 elements changed configuration.
Attribution is correlation, not causation. CAISO does not publish shift factors. We can show that an element binds disproportionately during a node's congested hours; we cannot show that curtailing a load there relieves it.
One site fails our own control. Rio Oso is the single node of 39 that does not beat a matched-random benchmark, because the element it was scored on stopped binding between years. We left it in the ranking rather than quietly removing it.
None of that is in the small print. It is on the console, in the deck, and in the regulatory comment, because a screening tool that hides where it breaks is worse than no screening tool.
Not a purchase. A test. Name five transmission nodes you have already studied; we write down what our screen says for those five and send it to you before you tell us anything, so the answer is committed rather than fitted. Then you score it against what you already know.
It costs about half a day of one engineer's time, no data leaves your building, and we publish the result either way — including if we are wrong.