Project Structure

The repo is a polyglot monorepo with a React frontend, a Rust workspace backend, a Bun-driven data pipeline, and a Next.js marketing/docs site.

Frontend conventions

• Components are organized by domain (broadcast/, panels/, controls/) rather than by atomic level (atoms/, molecules/), which makes it easier to find the place where the verse-overlay rendering lives or where the queue actually mutates.
• Hooks are thin and only own subscription/cleanup logic — the business logic lives in stores.
• Stores are Zustand slices, one per domain. Cross-store coordination happens through small, named helpers rather than global selectors, so a refactor stays local.

Backend conventions

• Each crate has a single owner concept (audio, STT, Bible, detection, broadcast, API). Adding a new sense — say, a transcription engine — means a new module in the existing crate, not a new crate.
• Tauri command handlers live in rhema-api. Other crates expose pure Rust surfaces and never import tauri::.

Data pipeline conventions

• Idempotent: every script in data/ checks for existing output and skips if it's already done. setup:all is the orchestrator that runs them in order.
• Reproducible: model versions and translation sources are pinned in script constants, not environment variables, so two operators running setup:all produce identical artifacts.