For my graduation project, I am building a hybrid system to detect fake AI-generated product images for e-commerce websites. The model uses ResNet-50 to find visual glitches and CLIP to find logical errors. I implemented a Patch Selection technique to focus on small details and textures that fake images usually get wrong.
Figure 1: Hybrid Model Architecture
Connect-4 Move Predictor (1st Place Winner)
Timeline: -
Developed an AI model to predict the best winning move in a Connect-4 game by combining two powerful models: XGBoost and CNNs. My team (Neurons) competed in a class-wide Kaggle competition to test the model on unseen game scenarios, and we successfully ranked 1st place on the private leaderboard with a score of 0.738.
Figure 2: 1st Place Ranking on the CSC462 Kaggle Private Leaderboard
