https://heidemann.me/https://heidemann.me/blog/2026-03-06-attention-monoid/https://heidemann.me/blog/2026-03-06-attention-monoid/A derivation of numerically stable softmax attention via a monoid structure underlying FlashAttention.https://heidemann.me/blog/2026-03-13-tools-that-work-together/https://heidemann.me/blog/2026-03-13-tools-that-work-together/An optimistic argument that the current AI wave may push software ecosystems toward better interoperability, automation, and composability.https://heidemann.me/blog/2026-03-21-mlir-was-so-close/https://heidemann.me/blog/2026-03-21-mlir-was-so-close/A critique of MLIR as a de facto compiler IR standard, from the perspective of implementation-independence, tooling, and dialect interoperability.