Girish Gupta

I am a Machine Learning Engineer with 2+ years of experience specializing in computer vision and MLOps, focused on turning research-quality models into reliable production systems. I’ve implemented and optimized training pipelines to reduce training overhead by 90% through distributed systems and optimization, while keeping reproducibility and deployment readiness front and center.

I strengthen real-world performance by enhancing dataset robustness using advanced augmentation and probabilistic inference. In production work, I developed a Dockerized environment for large-scale face recognition training (PartialFC), ensuring reproducibility (CUDA 11.3, PyTorch 1.12, MXNet) and cutting training time from 3 days to 5 hours per epoch.

I also build efficient, deployable vision models for challenging conditions—implementing head pose estimation via pruning and quantization to reduce model size by ~4x and inference timing by ~9x with minimal accuracy loss. I’m passionate about Explainable and Visual AI, leveraging state-of-the-art methods and research to solve real-world problems, from tracking pipelines to segmentation and geospatial deep learning.