About#

Hi, I’m V 🌸

I’m a Computer Science undergrad at UW-Madison, graduating in spring 2027 and heading into a master’s program afterwards. I’m fascinated by AI infrastructure: how to make large-model training run faster, and how to squeeze every last FLOP out of a GPU.

🛠️ What I work on#

Training performance — As an ML Infra intern at Kuaishou, I worked on MFU optimization for LLM supervised fine-tuning, FlashAttention integration, and performance tuning for MoE models on multi-GPU H200 clusters.
GPU kernels — I’m working through CUDA C++ and the CuTe DSL systematically, from reduce and softmax all the way up to GEMM and FlashAttention.
Open source — Contributing PRs to training-side AI infra projects.

📚 What you’ll find here#

Section	What’s in it
Learning	Notes on GPUs, CUDA and deep-learning systems — partly synced automatically from my notes repo
Thoughts	Ideas about tech, learning, and everything else
Life	Everyday notes, photos and short essays
Projects	My GitHub repositories, synced daily

📮 Find me#

GitHub: @vInThePrism

Built with Astro using the Firefly theme (based on Fuwari), hosted on GitHub Pages. Thanks for stopping by!