Pinned Loading

1. nano-sglang nano-sglang Public

   A tiny LLM inference engine, built from scratch to understand how the fast ones actually work. This is not a production serving stack and it is not trying to be. It is a from-scratch reimplementati…

   Python

2. kernel-labs-cuda kernel-labs-cuda Public

   CUDA kernel experiments for matrix multiplication, reductions, activations, and LayerNorm. Includes correctness checks and performance benchmarks. Built for hands-on GPU optimization.

   Cuda

3. Familiar-UI Familiar-UI Public

   Familiar UI detects a user's audience and generates a complete explanation, layout, and visual theme for that persona. The same topic can become a developer console, research paper, executive brief…

   TypeScript

4. foundation-models-pipeline-llmops-2 foundation-models-pipeline-llmops-2 Public

   End-to-end foundation model pipeline covering data collection, cleaning, preprocessing, training, and artifact management. Built to explore practical LLMOps workflows from raw data to reproducible …

   Jupyter Notebook

5. deep-learning-toolkit deep-learning-toolkit Public

   Deep-learning experiments with transformers, RNNs, LoRA, and Hugging Face. Includes GPU benchmarks and model comparisons. Built for hands-on learning.

   Python

6. systems-under-load systems-under-load Public

   Hands-on experiments in high-performance AI systems, from CUDA GEMM and memory optimization to MoE models and nano-SGLang. Focused on identifying bottlenecks, understanding system trade-offs, and i…

   Cuda