In this project, my team and I built an end-to-end waste classification system designed to support automated recycling workflows, transforming raw image data into real-time intelligent predictions. I contributed to developing the preprocessing and augmentation pipeline, balancing the dataset to 500 images per class to enhance generalization, and leveraged ResNet50 transfer learning to boost accuracy from 60% to 90%. We further optimized an RBF SVM for efficient inference and integrated OpenCV for live camera-based predictions.

In this project, my team and I developed a modular Java-based soft computing library aimed at transforming theoretical AI concepts into reusable and extensible software components. I contributed to implementing key features across three pluggable modules—Genetic Algorithms, Fuzzy Logic, and Neural Networks—designing flexible chromosome structures, fuzzy membership functions, inference engines, and configurable neural network layers.

In this project, my team and I built an intelligent Gomoku game to demonstrate how adversarial search algorithms drive strategic decision-making systems, supporting both Human vs AI and AI vs AI gameplay modes. I worked on implementing the Minimax algorithm enhanced with Alpha-Beta Pruning to optimize search efficiency while maintaining strong move evaluation, enabling real-time board updates and dynamic visualization using Tkinter.