Tech Stack: C, Linux Systems Programming, Namespaces, Cgroups, Networking Developed a Docker-like container runtime from scratch for lightweight OS-level virtualization Implemented process isolation using clone() with PID, Mount, and Network namespaces Applied Linux cgroups to enforce CPU and memory limits per container Built a virtual networking system using veth pairs and a Linux bridge Enabled isolated inter-container communication through custom network setup Configured NAT using iptables for secure outbound internet access Blocked inbound traffic to enhance container security Strengthened understanding of low-level OS concepts, networking, and resource management

Tech Stack: Python, PyTorch, Hugging Face Diffusers, CLIP, OpenCV, NumPy, Weights & Biases Built and evaluated generative models including GANs, DDPM, LDM, and SDXL on MNIST and Oxford Flowers-102 Implemented class-label conditioning for MNIST and CLIP-based text conditioning with Classifier-Free Guidance (CFG) Generated high-resolution, text-guided floral images using diffusion models Benchmarked optimizers (Adam, AdamW, Lion) to improve training performance Applied memory-efficient techniques like mixed precision (FP16/BF16), gradient accumulation, and LoRA Optimized training under limited VRAM constraints Evaluated model performance using FID and SSIM metrics Achieved stable training and improved visual quality of generated outputs