A mix of academic work and personal experiments. Built to learn, used to ship.
A deep learning system classifying ASL hand gestures using CNNs. Compared a lightweight baseline against VGG16/ResNet transfer learning on ~27,000 labelled images. The baseline achieved 100% validation accuracy in under 60 seconds.
A mobile app proposal tackling the cultural tourism engagement gap in Dubai. Features AI-curated heritage trails, AR historical overlays, and multilingual support — backed by qualitative interviews and survey research.
A controlled empirical study investigating whether neural agent architecture shapes emergent language structure in multi-agent referential games. Compares LSTM, GRU, Transformer, and MLP across 10 random seeds each.
A Russian-first LLM red-teaming platform and benchmark. Tests whether native-language adversarial prompts expose safety failures that translation-based multilingual testing misses — targeting YandexGPT, GigaChat, and open-source Russian-language models.
An interactive web application that visualizes common sorting algorithms in real-time, helping students understand how different algorithms work step by step.
A machine learning model that predicts student performance based on study habits and attendance. Achieved 87% accuracy using ensemble methods.
