I build what AI agents need to do real work in a web app: data primitives in Rust and WebAssembly, headless components agents can compose without breaking a design system, and architecture rules fast enough to run on every save. All open source. The data and UI layers ship MCP servers, so agents call them directly.
Volume 01
High-performance data primitives for browser visualization — Rust/WASM downsampling, binning, spatial indexing, and streaming aggregation, plus an MCP server for AI agents. It draws nothing. Feed it a million points, get back what matters.
Open source · MIT
npm install @vizcrush/core
A headless Web Components design system built with Lit: 25 unstyled, accessible components that work in any framework, plus a validated path for agent-generated UI and an MCP server for AI agents.
Open source · MIT
npm install @aetherui-kit/core @aetherui-kit/tokens
Fast dependency analysis and architecture rules for JavaScript and TypeScript, written in Rust on the oxc parser. Finds import cycles, enforces boundaries between parts of a codebase, and explores the dependency graph. Checks VS Code's source in under a second.
Open source · Apache 2.0
npm install --save-dev detangle
I built detangle, a Rust tool that checks import cycles and architecture rules on VS Code's source in 0.18 s. Here is how it was measured, how I checked it's correct, and the new ESLint rules.
I published AetherUI, a headless Web Component library with 25 core components and a validated path for agent-generated UI. Here is what the launch made me solve.
Rust's +simd128 produced 205 static SIMD instructions in my aggregate module, with none in the Welford loop I measured.
Start a project
pallav@debugdiary.dev