irhum khn

I've built and deployed deep learning systems for medical imaging, automation, and edge devices. At NECOP, I developed a DCGAN-based synthetic augmentation study for brain tumor MRI classification and deployed a 96% accurate classifier as a Dockerized FastAPI application on Hugging Face Spaces.

My medical-imaging pipeline covered preprocessing, DCGAN training, Swin Transformer fine-tuning, FID analysis, and statistical evaluation with PyTorch, Hugging Face Transformers, and scikit-learn. I also automated experiment tracking with n8n and authored an empirical study on DCGAN-based synthetic augmentation for brain tumor MRI classification.

At Software Product Strategies, I improved predictive-model accuracy by 15–20%, reduced experimentation time by 30%, and optimized edge models for 40% smaller size and 2× faster inference. At BAAM, I build AI-driven CRM automation workflows with n8n, Zapier, Gemini, and Claude.

My projects include a YOLO smart-home system running at 14–18 FPS on Raspberry Pi and a CNN/OpenCV driver-drowsiness detector running at 20–25 FPS. I enjoy turning rigorous ML experimentation into reliable, production-ready AI systems.