vercel-labs/json-render ? reverse-engineered prompt

I want to build a generative UI framework. The main idea is to let an AI generate user interfaces from natural language prompts, but in a safe and predictable way.

Instead of having the AI write raw code, it should be constrained to a catalog of predefined components and actions that I provide. The AI's output will be a JSON object that describes which of my components to use and how to arrange them. My framework will then take this JSON and render it.

This needs to be super flexible. Make it work with all the major front-end libraries: React, Vue, Svelte, and Solid. Also, let's add renderers for other targets too, like React Native for mobile apps, PDFs, HTML emails, and maybe even Remotion for video or React Three Fiber for 3D scenes.

To make it easy to get started, please include a pre-built set of components based on shadcn/ui for React. Let's structure the whole thing as a TypeScript monorepo.