Projects

Built an unconditional latent diffusion model on MNIST.

• Autoencoders + Channel Attention Blocks
• U-Net DDPM for denoising in latent space
• t-SNE embeddings & reconstruction quality

Real-time hand-gesture detection powering a fully automated game with cheating detection and dynamic winner celebration.

• YOLOv11 gestures at ~98.2% mAP@0.5
• Cheating detection: red mask & −1 penalty
• Winner recognition: animated crown overlay

Comparing reconstructions across latent sizes to illustrate the compression–fidelity trade-off.

Gaussian & salt-and-pepper noise removal via Attention U-Net + PatchGAN, achieving strong PSNR/SSIM.

Hands-on study of classic RL algorithms with Boltzmann exploration tweaks.

• DQN, SARSA, D3QN, and more

Multi-label classifier trained on 44k images for real-time tagging and dataset organization.

CNN-based app (Flask + JS) for accurate real-time recognition of Iranian celebrities.

LLaMA-3 + CrewAI pipeline for consistent, automated repository documentation.

Hands-on survey of supervised learning techniques on diverse datasets.

• KNN, SVC, Decision Trees, etc.