SIDE PROJECTS
Side opensource projects I do in low level programming in C++ and CUDA.
dvlt.cu
Suckless, single binary, zero-dependency CUDA/C++ inference engine for NVIDIA's DVLT. Reconstructs 3D scenes from a handful of images (depth + rays + camera pose => point cloud), no python, no torch, no framework.
frokenizer
Zero allocation, zero dependency, header only C++ BPE tokenizer for Qwen, using ahead-of-time DFA compilation to eliminate regex backtracking and heap overhead, reaching GBs/sec tokenization throughput.
qwen600.cu
+500 stars on github. Static, single batch inference engine for QWEN3-0.6B written in pure CUDA C/C++, no python dependencies. Faster than llama.cpp by ~8.5%.