Nandakishor M
I build small models that make decisions instead of writing essays about them.
Right now that means decision models: things that read a support ticket, a contract or an email and return a label and a probability, fast enough to sit in the request path and calibrated enough that you can branch on the number.
Writing
Vega: a decision model that rolls a ball down a hill
One prompt, one forward pass, five pooled vectors, and a small physics engine that settles into an answer with a probability you can use.
10 October 2026 · 9 min
How I built Laya
Keeping the reading half of a language model, throwing away the writing half, and training what is left to give honest probabilities.
9 October 2026 · 7 min