Viswanath Sai.

I'm Viswa - Viswanath Sai, on paper. I build software and companies, and I'm a founder backed by Entrepreneurs First. Most recently I co-founded Fettle, maintenance intelligence for the people who keep the physical world running.

I graduated from IIT (BHU), Varanasi in 2026, where I wrote a master's thesis on alloy optimization. Before founding, I spent my college years deep in open source - maintaining the Prometheus Operator and mentoring in Google Summer of Code.

A question I keep coming back to: whatever an LLM does is still driven by the next token - no external state changes that. But can we hold a state on top of what the LLM already does - an internal model of what it knows and doesn't know, an epistemic state - and use that abstraction to drive the LLM's behavior? I suspect durable reasoning, emotion and creativity live in that layer around the model, not inside it. Early signals and reading notes, not findings - the thinking-in-progress shows up in my essays.