Faisal E.

⭐️ Top Rated Plus - Top 1% of talent on Upwork 🕒 5000+ hours on Upwork ✅ 10+ years of Experience ✅ Excelling in delivering Fortune 500-scale projects Computer Vision Engineer for sports video and real-time tracking. Top Rated Plus, $250K+ earned on Upwork. I build player and ball tracking, pose estimation, and object detection that runs in production on the footage clients actually have: phone clips, CCTV, GoPro, broadcast. Most computer vision work dies between the notebook and the device. Mine doesn't. I'm currently the computer vision engineer on a sports analytics platform, turning game film into real-time insight for coaches and athletes. Before that: four months architecting an on-device photo intelligence engine in Core ML for a consumer AI app, and a real-time fall detection system written in C++ on NVIDIA Jetson. WHAT I BUILD Sports video analysis. Player tracking, ball tracking, pose estimation and biomechanics, action and event detection, jersey and team assignment, homography so tracked positions map onto the real pitch or court. Real-time video pipelines. Object detection and multi-object tracking with YOLO, OpenCV and PyTorch, stable across handheld phone footage, fixed CCTV, GoPro and variable lighting. People counting, zone and dwell logic, ALPR, OCR, event segmentation. Edge and on-device deployment. iOS with Core ML, Apple Vision and the Neural Engine. NVIDIA Jetson Xavier NX and Orin Nano with ONNX, TensorRT and CUDA. Model distillation and quantization to fit large vision models inside a real latency and battery budget without losing the accuracy that made them worth shipping. Production pipelines. AWS video processing at volume, batching and throttling, incremental processing, benchmarking, and documentation your engineers can build on after handoff. HOW I WORK I run your current pipeline on your own footage first and log exactly where it breaks, so we're working from a measured baseline instead of an opinion. Then targeted fixes, benchmarked against that baseline on the same clips. Then validation on a held-out set before anything scales. You get source code, models, training scripts, and documentation, not a demo nobody can reproduce. Stack: Python, C++, PyTorch, OpenCV, YOLO, TensorRT, ONNX, Core ML, Apple Vision, MediaPipe, CUDA, Docker, AWS. Good fit: you have video, a problem you can measure, and you care whether it runs on your hardware at your frame rate. Poor fit: pure LLM or chatbot builds, and projects where nobody can yet define what "working" means. Education: BS Computer Science, FAST NUCES. MS Computer Science (Data Science), Ryerson University. 🤝 Message me with your footage type, target hardware, and what "good enough" looks like as a number. If it's a fit, I'll tell you how I'd approach it. If it isn't, I'll say so.