Blog

GPU Kernels & Compilers: A Compressed Interview Course

From linear algebra to FlashAttention, MoE, and the XLA lowering pipeline — the working set for a GPU-kernel / ML-compiler interview, told with diagrams, the underlying math, and real kernels. Weighted toward the compiler / XLA / PTX axis.

Hacking LLDB for a great Zig debugging experience

Zig is better than C. And that’s great — until you try to debug it.

AI will write your next compiler!

It was a dark and stormy night — Well, a gloomy Kyiv morning.

Build your dreams!

I started coding over 20 years ago — writing BASIC on my lap during a 1.5-hour bus ride from Havana (where I went to school) back to Santa Cruz del

Artisanal Coding Is Dead, Long Live Artisanal Coding!

I recently discovered that I possess certain AI-assisted superpowers: I can now implement working features — the ones I actually need and want — far faster

Making OCaml recursive modules convenient

Jane St posted about recursive OCaml modules from recursive signatures a while ago.

Configuring Jujitsu (jj)

This is my Jujitsu (jj) repo config template. I usually symlink it to .jj/repo/config.toml and check it in.

Configuring Helix to auto-format Haskell code

This is how you configure Helix to auto-format Haskell code.

Zig type hackery and memory management

You can map and fold in Zig but lack of closures makes it unergonomic. Here's a completely silly example that shows Zig type hackery as well as memory