Computer Science Graduate. Focused on Reinforcement Learning.
-
Dissertation — RL Reward Function Design: Custom MuJoCo inverted pendulum, PPO/TD3 via Stable-Baselines3. Found reward curvature drives training stability; 100% eval survival. Repo →
-
DeepSeek-Instagram Automation Bot: Python system that integrates multiple APIs to automate social media workflows. ★ 2
-
World Bank Data Science Competition: XGBoost pipeline, ranked 339/1,322 (top 25%).
-
PixelStudio Pixel Art Editor: C++ system that integrates SFML and ImGui for real-time pixel art creation and editing.
-
Autonomous Mobile Robot (Webots, Python, SLAM): Developed an autonomous exploration system combining Bug2 navigation with frontier-based exploration. Built a real-time goal generation and recovery behaviours. Includes a discrete Bayes filter for localization under uncertainty
-
CNN Image Classifier (PyTorch, CIFAR-100): CNN (VGG-style blocks with BatchNorm + Dropout) on CIFAR-100, with a full pipeline dataset mean/std computation, data augmentation, train/val split, and MultiStepLR scheduling.
Studying the full language model pipeline by building one. Data curation → continued pretraining → post-training with SFT and DPO, applied to Arabic writing.
