Built an AI that spots fake images and explains why — even in blurry 32×32 pixels. Combined smart CNNs with a vision-language model to detect artifacts and reveal the truth behind AI-generated visuals.

Leveraged Deep Q-Networks in PyTorch and TensorFlow to train Unity-based football agents from scratch, achieving a 40% boost in strategic decision-making accuracy and an 85% success rate in mastering complex gameplay strategies. A 3D Unity simulation visualizes learned agent behaviors and tactics.

Built a high-efficiency AI agent for domain-specific question answering using GPT-4, combining multi-task learning with prompt engineering and CoT reasoning to streamline support workflows, cut response times, and deliver scalable cost savings.

DeepFusion-C is a hand-optimized C implementation of a four-layer CNN for MNIST digit recognition, leveraging SIMD intrinsics and OpenMP to deliver sub-millisecond, >99.2%-accurate inference with detailed latency, throughput and memory profiling.