A developer shares their experience building an anime recommendation engine using SvelteKit and TensorFlow.JS, coining the term 'middle-end' development. By keeping everything in a single SvelteKit app, they reuse code across data collection, model training (done in-browser via WebGL), and live inference (via tfjs-node in a worker thread). The architecture collapses what would normally be separate frontend, backend, and ML services into one Docker container, dramatically simplifying deployment and maintenance. The post covers the benefits (shared types, caching, code reuse, fast dev startup) and honest tradeoffs (poor team scalability, tightly coupled complexity) of this approach.
Sort: