Najmus Saqib

• Engineered DL models for a video analysis system, reducing false positives and missed detections by 10-15% over legacy models. • Optimized model deployment with GStreamer, NVIDIA DeepStream, and TensorRT, enabling real-time inference on 30+ video streams at 30+ FPS. • Built MLOps pipelines with MLflow for experiment tracking and DVC for dataset versioning/pipeline orchestration, accelerating training and deployment cycles. • Deployed microservices with Kubernetes for real-time inference delivery.

• Developed an illegal building recognition system on a dataset of 50,000 satellite images, achieving 90% precision and 75% recall. • Enhanced model interpretability by integrating Grad-CAM and SHAP explanations, providing stakeholders with transparent insights into model decisions for regulatory compliance. • Implemented a custom loss function to address the class imbalance and optimize for high precision. • Scaled training jobs using PyTorch's DDP across multiple GPUs, reducing training time by 40%.

Coursework: Machine Learning, Deep Learning, Computer Vision, Cloud Computing, Big Data Computing, Distributed Systems, Autonomous Computing, and Information Systems.