Intelligent systems.
Seen from within.
I’m Navish Kumar, a machine-learning researcher and systems builder at the University of Basel.
I study how systems change—from mathematical structure and continual learning to evidence-grounded agents and persistent worlds.
Three research objects: gain-graph structure, learning geometry, and a persistent laboratory.
Different questions.
A connected body of work.
My early work made local relationships measurable. During the PhD, the question became how a useful system can adapt—and how to see what that change costs.
Open the work atlasInteraction dynamics
What does an isolated message leave out?
Gain graph structure
What happens when one relationship stops agreeing?
Spectral certificates
Can the spectrum tell us how much repair a graph needs?
Urban micro-regions
Why does the better delivery vehicle change a few streets away?
Optimization geometry
Can a change of representation make learning easier to explain?
Experience Replay
Which memories actually counter forgetting?
Rank Feasibility
What if the correction does not fit inside the model?
Temporal replay value
When does memory become inertia?
CasePath
Should a plausible answer be allowed to become an action?
Persistent worlds
What changes when your next sentence edits the same world?
From an idea to something
you can actually use.
Mathematical models, experiments, and working interfaces—each with its evidence and its limits left visible.