Bowen Cheng https://bowencheng.dev/ Writing by Bowen Cheng on systems engineering, AI infrastructure, and technical projects. en-us Anatomy of a Tokenizer from First Principles https://bowencheng.dev/blogs/tokenizer https://bowencheng.dev/blogs/tokenizer Mon, 01 Jun 2026 00:00:00 +0000 Building a modern tokenizer from first principles: why Transformers need integers, and how Unicode, words, UTF-8 bytes, and Byte Pair Encoding each solve one piece of the problem. Writing SGEMM Kernels in CUDA https://bowencheng.dev/blogs/matmul https://bowencheng.dev/blogs/matmul Fri, 01 May 2026 00:00:00 +0000 CUDA SGEMM kernel notes covering a cuBLAS reference, naive implementation, shared-memory tiling, register tiling, and benchmark results. Building a Performant Compiler from Scratch https://bowencheng.dev/blogs/compiler https://bowencheng.dev/blogs/compiler Sun, 01 Mar 2026 00:00:00 +0000 Architecture, algorithms, and optimization techniques behind a 10k+ line compiler pipeline that lowers a C-like language to x86-64. I.C.U. https://bowencheng.dev/blogs/icu https://bowencheng.dev/blogs/icu Mon, 01 Dec 2025 00:00:00 +0000 Building I.C.U. (Intrusive Camera Unit), 3D printed eyes that track your face using a microcontroller and OpenCV. Splashv2: Automating Large-Scale Traceroute Ingestion https://bowencheng.dev/blogs/splash https://bowencheng.dev/blogs/splash Sun, 01 Jun 2025 00:00:00 +0000 A scalable system for automating traceroute ingestion and analysis, built for Aqualab at Northwestern University.