Senior Software Engineer · Computer Architect
Hi, I'm Kyushick Lee.
I build kernels, runtimes, and LLM serving stacks for AI accelerators at Microsoft — with a background in computer architecture, GPUs, and system resilience.
Currently exploring
LLM inference & trainingAI accelerators / MaiaKernel authoring & MoE kernelsPyTorch / ONNX Runtime / TritonSystems performanceResilience & checkpointing
Selected projects
All projects →Vectorizing 64-bit Integer Division
Emulating 64-bit integer division using 32-bit SIMD lanes for a measurable speedup.
active SIMDInteger arithmeticPerformance
Building PyTorch from Source
A reproducible workflow for building PyTorch from source with CUDA on WSL2.
maintained PyTorchCUDABuild systems
Selected writing
All writing →- The Question I Never Asked My brother is an architect. His question about what it means to be one led me to ask the same of myself as a computer architect.
- Why I Built This Site A research-engineer's notebook: portfolio, writing, and an interconnected digital garden.
Research notes
All research →- Validating Numerical Precision for Low-Precision Ops Methods and tools to detect and validate numerical precision issues when implementing math ops in low-precision types like bfloat16.
- Numerical Precision Challenges in ATen Operations Precision-loss pitfalls in low-precision (bf16/fp16) PyTorch ATen ops — catastrophic cancellation, absorption, ULP, and how to reason about them.
- LLM Inference Optimization Where the time and memory actually go when serving large language models.
- Barrett Reduction Replacing division by a constant modulus with a multiply and a shift.
- Knuth Algorithm D Schoolbook long division on machine-word limbs, done carefully.
Contact
Interested in systems performance, AI infrastructure, or low-level optimization? Reach out via email or GitHub.