moiz khan

I'm building and validating computer vision and generative AI systems at the University of Oulu's Wireless Communication Engineering Group. My Vision Transformer image-quality pipeline improved evaluation accuracy from 63% to 95% through systematic testing.

I fine-tuned Stable Diffusion 3.5-M with multi-head attention for generative image tasks and engineered a denoising optimization that cut training time by 31% while preserving output quality. I also implemented Flow-GRPO for controllable image generation and trained across multi-node GPUs on Finland's national CSC supercomputer.

Alongside research, I build Python automation pipelines for large-scale data processing through Upwork. My projects include real-time video object tracking, retinal disease classification, and edge semantic segmentation.