Synthetic Data for Model Evaluation

Investigated how synthetic test data generated by CTGAN and TVAE can improve ML model evaluation under distributional shift. Proposed the 3S-Testing framework and compared generative approaches across MLP, Gradient-Boosted Trees, and Random Forest.

Python CTGAN TVAE ML Evaluation
Read Paper → GitHub →

LuAI: LLMs that Teach Lua

Fine-tuned instruction-based LLMs (Qwen3-4B, CodeLlama-7B) to teach Lua programming step-by-step. Used model distillation to generate synthetic training data and prompt engineering to produce structured, educational code explanations.

LLMs Fine-Tuning Distillation Lua
Read Paper →

CIFAR-10 Image Classification

Built and trained multiple CNN architectures on CIFAR-10, iterating on batch normalization, pooling strategies, and optimizer tuning. Implemented Deep Dream visualizations to explore what convolutional filters learn at each layer.

PyTorch CNNs Deep Dream CIFAR-10

Digital Carcassonne

Desktop implementation of the board game Carcassonne in C++ with a 4-person team. Built modular game-state management, tile placement logic, and a full UI using SFML libraries with professional Git workflow.

C++ SFML Game Dev Team Project