I benchmarked caveman against two words
Caveman, a popular Claude Code compression plugin, vs. "be brief." 24 prompts, six categories, five arms. The two-word prompt matched it on tokens and quality.
I work on AI evals and reliability for agentic systems. Skeptical of tooling hype, mostly because I keep measuring it.
Caveman, a popular Claude Code compression plugin, vs. "be brief." 24 prompts, six categories, five arms. The two-word prompt matched it on tokens and quality.
Building a simple REPL/shell in Rust with Tokio and anyhow — parsing input into typed commands and implementing echo, ls, pwd, cd, touch, rm, and cat.
Cloning the Apple calculator with Tauri, Yew, and Tailwind. Operator precedence, decimals, and a UI close enough to fool muscle memory.
Building a desktop Pomodoro timer in Rust end-to-end with Tauri, Yew, and Tailwind — system tray, state management, and a controls UI.