Kh Safkat Amin

Developed a modular Python framework for harmonizing multimodal trajectory datasets, with structured storage in a MongoDB backend. Automated the enrichment pipeline by integrating external data sources, including OSM, HD maps, and DWD weather APIs. Developed analytical modules for junction detection, maneuver classification, interaction labeling, and computation of risk metrics. Designed advanced MongoDB queries enabling efficient scenario-based filtering and data retrieval through application interface.

- Operated a tribological test bench to collect experimental data on wear and boundary layer formation. - Developed and validated predictive models using classical and neural network-based techniques implemented in scikit-learn and TensorFlow to analyze complex tribological phenomena.

Upgraded perception from basic lane detection to a BEV-based solution using IPM and SegNet for semantic segmentation on Jetson AGX Xavier with ROS 2. Trained YOLO models for traffic sign detection using transfer learning and custom augmentation. Built datasets from ROSBAGs, annotated with Roboflow, and deployed full pipelines. Validated in Gazebo and real-world tests. Contributed to Jetson setup, debugging, and documentation in an Agile team.

Worked at RAIL, a Bangladesh automotive assembly company specializing in CKD passenger and commercial vehicles. Supervised quality control and repair operations, ensuring regulatory compliance and performing diagnostics, root cause analysis, and FMEA. Led Material Management, optimizing logistics and digitalizing workflows with advanced Excel to enhance efficiency and traceability.