Ahmed Ibrahim

•Architected an Azure-based Retrieval-Augmented Generation (RAG) system for automated information extraction, orchestrating the end-to-end data pipeline including cleaning, chunking, and vector database indexing. •Executed benchmarking and validation of Large Language Models via advanced prompt engineering, minimising hallucinations and ensuring strict factual precision. •Engineered and deployed deep learning detection systems using MMDetection for automated aerial imagery analysis, focusing heavily on high

•Topic: Improving Semantic Segmentation Models through Layout Preserving Synthetic Training Data Generation. •Deployed Stable Diffusion XL (SDXL) and ControlNet to generate layout-accurate synthetic imagery, conditioned on Cityscapes segmentation masks to achieve high visual variance (weather, lighting, textures) without manual labelling. •Achieved a demonstrable increase in mean Intersection over Union (mIoU) for complex object classes, successfully resolving data scarcity issues through automated dataset

•Developed image classification models and automated data processing pipelines, implementing label refinement and regularisation strategies to mitigate model overfitting. •Deployed an optimised inference pipeline onto resource-constrained edge hardware, achieving a real- time model accuracy of 95% in production. •Programmed OpenCV-based computer vision solutions tailored for industrial hardware, successfully automating object and pattern recognition processes. Tech Stack: Python, Deep Learning, Edge-AI, Op

•Optimised a TensorFlow/Keras image classification model for crash-test analysis, executing data validation to enhance predictive accuracy. •Automated internal search workflows by developing a custom Python toolkit, accelerating retrieval processes by 90% and digitising cross-department workflows using Power Automate and SharePoint. Tech Stack: Python, TensorFlow, Keras, Microsoft PowerAutomate.