Projects

A deep reinforcement learning framework using Double Deep Q-Networks (DQN) to dynamically optimize traffic light timing. The system learns to reduce congestion and waiting time through environment feedback and adaptive decision-making.

A retrieval-augmented generation (RAG) system that integrates external document evidence with large language models to produce accurate and context-aware answers. It demonstrates how combining semantic search with generation improves factual reliability in AI systems.

A neural architecture search (NAS) project focused on designing compact convolutional neural networks (CNNs) with high performance and low computational cost. It automates architecture discovery for deployment on edge and mobile devices.