Hi, I'm Md Shakhaout Hossain

As project lead, I directed the design and execution of a system to track individuals across non-overlapping cameras, managing the full project lifecycle.

Developed a video application that replaces faces for privacy. Instead of jarring blurs, our system intelligently generates a new face that matches the original person's movements and expressions.

Engineered and deployed an optimized pose estimation model on a resource-constrained NVIDIA Jetson device, managing the full data and training pipeline.

Managed the end-to-end data pipeline, from quadcopter image acquisition to training a YOLOv3 model for automated vehicle detection.

Engineered automated pipelines on AWS for fraud detection and traffic forecasting models, which have run in production for 3+ years.

Built a real-time, multi-channel streaming app on NVIDIA Jetson, synchronizing video and audio with ultra-low latency using UDP for a telemedicine use-case.

Engineered a classification model to differentiate throughput rates in industrial equipment for a leading machine tool manufacturer.

Developed regression models from large-scale sensor data to predict potential structural failures in railway tracks.

Trained an AutoEncoder model to automatically detect defects in solar panels from aerial video, developing the full data pipeline.

Researched and applied advanced prompting techniques (e.g., CoT, ReAct) to improve the reliability and task-specificity of Large Language Models.

Implemented a complex research paper to estimate human Center of Mass (COM) and joint torques using an HRNet model from standard 2D video.