Latest note: Rewriting wc in Rust
About Me
These days I build tools at Factory AI that let agents write and ship code. Watching an AI spin up a full feature while you grab coffee still feels surreal.
Away from work, I tinker with my homelab. Self-hosting gets addictive fast—media servers, home automation, whatever open source tool I stumbled across on GitHub at 2am.
I blog about the mess. Lately that's meant rewriting open source apps in Rust and Go, publishing the benchmarks, and documenting what mission mode actually shipped.
Projects
Things I've built, broken, and fixed. Mostly self-hosting tools and benchmarks.
parser-go
Go HTTP service for parsing web logs, built in one Factory mission-mode session with parity-gated benchmarks.
k8s-homelab
Production-like Kubernetes homelab with Talos, Ceph, GitOps, and autonomous SRE workflows.
xxhash-rs
Clean-room Rust reimplementation of xxHash with NEON-accelerated XXH3 and 508/508 parity tests.
the_silver_searcher
Incremental Rust rewrite of the Silver Searcher (ag) with published parity tooling, manifests, and a dedicated rust-ag workspace.
wc-rs
AVX2 SIMD rewrite of GNU wc. 27x faster on default mode, 21x on char counting, with rayon multi-file parallelism.
Blog
I write about breaking things, fixing them, and measuring the results.
Rewriting wc in Rust
A Rust rewrite of GNU wc that uses AVX2 SIMD to count words at 7.5 GB/s -- 27x faster than GNU coreutils on the default mode, discovered through 12 systematic experiments where branchless scalar was a dead end and SIMD was the only path past 1 GB/s.
Single-file throughput (MB/s) on a 100MB generated text corpus across default, line-only, word-only, char-only, and byte-only modes. Multi-file throughput on 1,000 files (~53 MB total). Three-way comparison against GNU wc and cw (Freaky/cw).
Rewriting a Python web log parser in Go
I rewrote a Python web log parser as a Go HTTP service. On 1.89 million lines of real NASA access logs, the Go version parses 3.3x faster than the Python baseline. A parity harness verifies both implementations produce identical output before allowing performance claims.
Head-to-head on 1.89M lines of real NASA Kennedy Space Center access logs (July 1995). Go parses 3.3x faster than Python.
Start with the latest post, then browse the full archive when you want the rest.
View all posts