Muhammad A.

Most AI models perform well in training but break down in production; lighting changes, camera movement, low FPS, and noisy streams destroy accuracy. I build Computer Vision and Edge AI systems that actually hold up in the real world. What I do: I handle the full pipeline, dataset prep, model training, optimization, and real-time deployment for object detection, tracking, segmentation, pose estimation, OCR, and video analytics. I specialize in deploying these models on edge devices (NVIDIA Jetson, embedded ARM systems) and optimizing them for low-latency, high-FPS inference using TensorRT, ONNX, and CUDA. Tech stack: Python, C++, OpenCV, PyTorch, TensorFlow, YOLOv5/v8/v11, Ultralytics, ONNX Runtime, TensorRT, CUDA, NVIDIA DeepStream, FastAPI/Flask Recent use cases: CCTV surveillance & video analytics, industrial/manufacturing inspection, ANPR & license plate recognition, retail analytics, traffic & smart parking systems, drone vision, sports analytics Where I deploy: Jetson Nano/Xavier/Orin, Raspberry Pi, embedded edge devices, Linux servers, cloud GPU infrastructure If you're scaling a CV product, optimizing a YOLO pipeline, or need a model that actually performs outside the lab, let's talk about your project.