I am 21, exploring the frontiers of Embodied Cognition — closing the sim-to-real gap so that what a model learns in simulation holds up in the real world. At eyecandy robotics, I take learned policies from training all the way onto real hardware.
I specialize in taking research built for the cloud and making it run at individual scale — shrinking training and inference onto a single GPU and far smaller datasets, reproducing frontier papers end to end, and squeezing real gains out of them through careful tuning and finetuning.
I handle version control as @5usu, ship hot takes as @sujalDweD, and resolve exceptions via @sujaldwed@gmail.com.