At LEXIT, I built a handwritten digit recognition CNN in Python and PyTorch, handling data preprocessing, model training, and evaluation.
I also built an instrument sound classifier and deployed it with Docker and FastAPI for real-time predictions. For multilingual profile matching, I used sentence transformer models to match Farsi and English profiles by skills and interests.
I developed a virtual try-on system using transformer models to map clothing onto user images, with image preprocessing and model-based alignment. I’m interested in practical machine learning and AI systems, including reproducible workflows and production-style projects.
