Hi, I'm Saman.

I'm an ML engineer at Meta working on LLM efficiency: quantization, inference optimization, and the GPU kernels underneath. Before that I spent a decade on backend and infrastructure work, from SRE and multi-cloud architecture to co-founding a trading company's software stack.

This site collects my writing. The blog covers how modern LLMs actually work under the hood (attention, MoE, RoPE, sampling, batching) and a hands-on series on writing Triton and CUDA kernels, with benchmarks. I also build MintEngine, an educational inference engine for checking model implementations layer by layer against production engines.

Recent posts

• 
  RoPE: The Phase Geometry Behind Long-Context Transformers

  How rotating query and key vectors turns absolute position into relative attention

  Jun 6, 2026 · #llm
• 
  From Greedy to Nucleus, How LLMs Choose the Next Token

  Decoding algorithms and sampling techniques explained

  Dec 22, 2025 · #llm
• 
  Matmul

  A tiled CUDA matmul kernel with shared-memory caching, benchmarked against Torch.

  Dec 1, 2025 · #kernels
• 
  2D Workloads

  Mapping images onto CUDA thread blocks: row-major memory, 2D indexing, and a grayscale conversion kernel.

  Nov 22, 2025 · #kernels
• 
  Fused Softmax::P3::Cuda Kernel

  A CUDA softmax kernel taken from a naive per-row loop to shared memory, float4 loads, and warp-shuffle reductions.

  Nov 16, 2025 · #kernels
• 
  Fused Softmax::P2::Triton optimization

  Debugging a Triton softmax that falls off a cliff at large row sizes, and what block size has to do with it.

  Nov 9, 2025 · #kernels

All posts →