Eshaan Gupta

At Terafac Technologies, I lead production 6-DoF pose estimation for robotic weld-seam inspection, combining UNet++ segmentation, ICP refinement, and multi-view fusion on deployed robot hardware.

I reduced mean reprojection error from 14 px to under 3 px and achieved stable on-robot accuracy of 2–3 mm in Z, 5 mm in X, and under 1° rotation. I also built a VGGT and SegFormer masking model that generalises zero-shot to unseen industrial parts, replacing per-part retraining.

I train and deploy deep-learning models with PyTorch, GCP, ONNX, and Docker, and have shipped MERN applications, an LSTM grid-frequency forecaster, and a diabetic-retinopathy classification model published at ICMLSC’26.