Kavya Sree Kammari

I got into AI because I watched it transform the world during COVID. That was my cue.

Since then I learn by building — a home robot that moves and reasons, agentic pipelines, RAG systems at scale. If I can’t make it work in the real world, I don’t really understand it yet.

My obsession lives in the gap between impressive demos and real deployments.

A benchmark score tells you it can pass a test. I want to know what happens after — in the wild, with messy data, real users, and edge cases nobody thought to write down.

I build for embodied intelligence. Robots included.

Robots are genuinely cool — yes, that’s reason enough. But also, the hardest AI problems show up when your model has to act in a physical space it can’t fully see. MAVRIM taught me that.

I care about interpretability as much as capability.

Training a model that scores well is the easy part. Understanding why it makes a decision, and where it will fail, is where the real work is. Not because it's required, but because it's what separates research from engineering theatre.

I’d rather understand why it decided.