A UMD-forged PhD who taught meshes to feel pain via backpropagation, publishing DMesh at NeurIPS while databases and Docker remain forbidden runes he refuses to speak.
“Solid ML with actual NeurIPS and SIGGRAPH papers — DMesh, GIPPO, differentiable everything. You're doing real research while the rest of us fine-tune prompts and call it work.”
“Heavy backend evidence, but it's all research pipelines and CUDA kernels — you can differentiate a mesh but have never once handled a 500 error at 3am.”
“Decent signal from complex differentiable sim architectures, but your systems scale to 'one GPU and a deadline' rather than a million users.”
Zero database signal — your 'persistence layer' is a .pkl file named final_final_v2 and a prayer to the filesystem gods.