// 001
currently building descrai.io ↗
Adithya
Kumar
Founder · Mathematician
Find a hard problem. Break the assumptions. Build the path forward.
// 002
Descrai
THE PROBLEM
50% of published research is irreproducible.
Not because scientists are careless. No tool exists to capture what actually happens at the bench. The raw process data is lost the moment the experiment ends. Hypotheses, protocols, deviations, observations: gone. The loop between experiment and analysis never closes, and AI science can't advance beyond the quality of data that fuels it.
THE SOLUTION
Descrai is the capture layer: a VLA pipeline that automatically takes experimental notes and verifies protocol adherence in real time, paired with smart safety glasses that give the system eyes and ears at the bench. It closes the data gap that any future AI scientist needs to work from.
// 003
Work
Descrai
Descrai v3: Lab Intelligence
50% of published research is irreproducible, largely because the raw process data is never captured. Descrai v3 is a VLA pipeline that automatically takes experimental notes and verifies protocol adherence in real time. To get the technology into the lab, we built smart safety glasses that give the system eyes and ears at the bench, bridging the data gap that any future AI scientist will need to work from.
Selected work
Lasserre Hierarchies for MAX-CUT
Dissertation: deriving and comparing Lasserre SDP relaxation hierarchies across multiple formulations of the MAX-CUT problem. Investigates how tightly successive levels of the hierarchy bound the optimum and where different formulations diverge in the quality and cost of their relaxations.