- TorchLite: My open-source neural network library with a PyTorch-like API, built on top of a scalar-valued autograd engine. (Project Link)
- GPT-2 from Scratch: Implemented GPT-2 entirely from scratch using Python without any loops, making it 20x faster (for 3 transformer layers). (Project Link)
- File System: Implemented an efficient, multi-threaded file system in C++ as part of the Operating Systems course project. Our system was 8x faster than Python implementations on large files. (Project Link)
- Project Photon: Built a Google Photos clone using MERN stack, deployed on Google Cloud Platform with a microservices architecture. (Project Link)
- National Open-Courseware: Built a clone of MIT OpenCourseWare for Pakistani universities, adding AI-based translation, chat rooms, and a digital library. (Project Link)
- GAN Architectures from Scratch: Implemented six GAN variants (DCGAN, CycleGAN, ConditionalGAN, SNGAN, WGAN-GD, and StyleGAN) in Python and Numpy. (Project Link)
For more projects, visit my GitHub profile.