Member of technical staff · Inference & optimizations · Reactor
Generative models, down to the kernel.
I work at Reactor on world models. Before that I was at Meta working on LLM efficiency: quantization, inference optimization, and the GPU kernels underneath, and before that I spent a decade on backend and infrastructure, from SRE and multi-cloud architecture to co-founding a trading company's software stack.
I also build MintEngine, an educational inference engine for checking model implementations layer by layer against production engines.
Recent writing
All 19 posts →- RoPE: The Phase Geometry Behind Long-Context TransformersLLMs
- From Greedy to Nucleus, How LLMs Choose the Next TokenLLMs
- MatmulGPU kernels
- 2D WorkloadsGPU kernels
- Fused Softmax::P3::Cuda KernelGPU kernels
- Fused Softmax::P2::Triton optimizationGPU kernels
- Fused Softmax::P1::Naive & Triton ImplementationGPU kernels
- Vector Addition::P4::OptimizingGPU kernels
Publications
- Gemma-3: Architecture and Mathematical FoundationsTechnical report · 2025
- Benchmarking Efficiency Techniques in GenAI Foundation Models Using an Elo-Based Performance Evaluation FrameworkMLArchSys workshop at ISCA · 2025