How martinjm97/ennui works
Project Type and Purpose
• ENNUI is a frontend-only, single-page application (SPA) for designing, training, visualizing, and exporting neural networks directly in the browser.
• Users drag-and-drop layers and activations to build network architectures, run training loops on in-browser data (e.g. MNIST/CIFAR-10), watch live training metrics, and export equivalent code in Python or Julia.
File Structure Analysis
• Top-level:
• Configuration and build: webpack.config.js, tsconfig.json, tslint.json, package.json, npm-shrinkwrap.json, build_prod.sh
• Distribution output: dist/
• src/
• src/model/ – “backend” logic in the browser: network construction, graph algorithms, in-browser training loops, parameter management, code generation (Python & Julia skeletons), export and save-state routing.
Insights from README
• Clear separation of concerns: UI vs. model logic.
• No server/backend: all computation, storage (URL state), and code generation happen client-side.
• Build steps via npm/webpack; one command bundles TS + SCSS into dist/.