👋 Hi, I'm Syed Ahad Ali, an AI Inference & Video Systems Engineer specializing in high-performance computer vision pipelines, hardware-accelerated video decoding (NVDEC), and optimized model deployment (TensorRT).
I focus on bridging the gap between research and production by porting Python prototypes to heavily optimized C++/CUDA architectures, ensuring maximum throughput and minimal latency for real-time video analytics.
- High-Performance Inference: TensorRT, PyTorch, OpenCV DNN, C++ Model Deployment
- Video Processing & Hardware Acceleration: NVIDIA Video Codec SDK (NVDEC), OpenCV, Parallel Computing
- Tracking & Analytics: DeepSORT, ByteTrack, BotSort, YOLO Architectures
- Systems Engineering: Python-to-C++ Porting, Profiling, GPU Memory Optimization
Parallel Video Decoding Architecture
- Designed a benchmarking suite for parallel video flow architectures leveraging hardware-accelerated NVIDIA Decoders (NVDEC).
- Focused on concurrent stream processing to maximize GPU utilization and minimize latency in upstream deep learning pipelines.
Hardware Decoder Validation & Benchmarking
- Developed a rigorous validation pipeline to guarantee accuracy when migrating from CPU-based decoding (OpenCV) to GPU hardware decoding (NVDEC).
- Automated tensor extraction and comparison to ensure pixel and tensor parity across drastically different rendering and decoding pathways.
Real-time Safety Monitoring System Optimization
- Engineered and optimized a real-time tracking pipeline moving from a Python/PyTorch baseline to a highly accelerated C++/TensorRT architecture.
- Integrated multi-object tracking (ByteTrack/BotSort) with custom optimizations to maximize FPS on edge and server-class GPUs.
- Profiled and identified pipeline bottlenecks, transitioning seamlessly from OpenCV DNN to TensorRT natively in C++.
