Srijan Ratrey
I build and evaluate ML systems — content safety classifiers, retrieval pipelines, and the measurement harnesses that tell you whether any of it actually works.
Macro-F10.684
Accuracy trap89.83%
N test23,936
Most of what I find interesting sits in the gap between a model that scores well and a model that is useful. A classifier that predicts nothing at all can post 89.8% accuracy; a retrieval system can return confident citations for the wrong document. So the work I care about is choosing the metric, calibrating the operating point, and being specific about where a system fails.