Mostafa Abdelraheem

Designed and implemented a dynamic semantic map-updating system for mobile robots using open-vocabulary 3D scene understanding. Deployed and evaluated the full pipeline on a Waveshare UGV Rover with Jetson Orin Nano and Intel RealSense D455 in real indoor environments.

Developed real-time vehicle localization and mapping visualization tools using Mapbox GL (frontend: Angular, backend: Python). Upgraded object detection pipeline from YOLOv5 to YOLOv8, improving accuracy for autonomous driving research. Deployed YOLOv8/YOLOv11 on Qualcomm QRB5165 using the SNPE SDK with INT8 quantization, significantly reducing inference latency.

Designed and implemented a quality control scheme for validating the bounding box annotations from outsourced labeling companies. Used FiftyOne and CVAT to visualize, compare, and monitor annotation quality across multiple labeling sources.

Built responsive UIs for an e-commerce platform using ReactJS and Redux. Developed shop owner dashboards with integrated analytics to track performance.

Thesis: Open-Vocabulary Semantic Mapping with LLMs and VLMs (Chair of NeuroRobotics)