Ahmad Liaqat

Real-Time Background Replacement System: Developed GPU-accelerated web app for video background removal/replacement using BiRefNet segmentation models. Implemented parallel frame processing (ThreadPoolExecutor) handling 200+ frames with real-time streaming preview. Built multi-background support (color/image/video) with MoviePy video editing and PyTorch CUDA inference. Deployed on Hugging Face Spaces with dynamic GPU/CPU fallback and configurable performance optimizations. Tech Stack: Gradio, Transformers, PyTorch, MoviePy, PIL, NumPy, concurrent.futures Real-Time Biometric Eye Segmentation & AR Virtual Try-On: Developed an end-to-end computer vision pipeline for highprecision Iris and Pupil segmentation using YOLOv8 and MediaPipe Face Mesh. Engineered a Hybrid Tracking System that improved performance by 30% through selective segmentation intervals and temporal mask interpolation. Implemented Kalman Filters and EMA Smoothing to stabilize virtual contact lens placement, achieving a jitter-free AR experience at 30+ FPS. Managed the full ML lifecycle: from dataset class distribution analysis and data augmentation to model training with PyTorch and real-time webcam deployment.

FlipperPro - Real Estate Investment Management Platform: Developed a comprehensive SaaS platform for real estate investors to manage house flipping and rental property investments. The application enables users to track investment deals, analyze profitability using industry-standard calculations (Flip & BRRRR analysis), manage project budgets, track expenses, organize tasks, and generate financial reports. Frontend Technologies: React 18, React Router, Tailwind CSS, Context API, Axios, Google Maps API, jsPDF, Victory Charts, React Icons, React Testing Library, Jest Backend Technologies: Node.js, Express.js, MongoDB, Mongoose, MySQL, Sequelize, JWT Authentication, bcryptjs, Multer, CORS