Nandan Patel

I’m a results-driven AI/ML Engineer and postgraduate Computer Engineering student with hands-on experience across computer vision, generative models, and agentic AI systems. I’m focused on turning models into production-grade, reliable solutions that improve accuracy and reduce operational effort.

At ISRO Space Applications Centre, I researched synthetic PCB defect image generation using Diffusion Models, cTransGAN, and SAGAN, benchmarking architectures on the DeepPCB dataset. I also built a PCB structure detection system using SAM, Canny edge detection, and skeletonization to enable precise structural boundary analysis and reduce manual inspection effort.

As a Full Stack AI Engineer (freelance), I integrated Gemini OCR into an AP Invoice Bot for part number extraction and document classification, improving extraction accuracy and cutting manual data entry time by 80%. I delivered end-to-end AI-assisted automation pipelines for invoice validation, PO matching, and Infor IDM uploads, processing 200 invoices per month with a 20% reduction in processing errors/time.

My broader work includes a dual-subnetwork NILM architecture with self-attention and dilated convolutions for appliance-level energy disaggregation, plus research in peer-reviewed publication for multi-class skin lesion classification using ResNet50. I’m energized by challenging, real-world problems—especially where robust ML meets clean engineering.