ModelForge

Problem

Turning JSON into typed models usually means a cloud service or a language model: the result changes between runs, and the data, the field names, and the code all leave the machine.

Engineering

The architecture is JSON → a universal intermediate representation → per-language generators, in a pnpm monorepo. The IR is inspectable and editable before generation, the generators are decoupled, compatibility is declared in explicit profiles per Java and Spring Boot version, and everything runs in the browser: no backend processes the data.

Outcome

An MVP deployed to production on Cloudflare Workers, with an interface in eight languages and automated tests across Chromium, Firefox, and WebKit.

Capabilities

• Converts JSON to TypeScript, Java, Spring Boot, C#, and Python.
• Deterministic generation: the same input and settings always produce the same output.
• Infers types, nested structures, nullability, and required fields.
• The intermediate model can be edited before the code is generated.
• Compatibility profiles for different Spring Boot and Java versions.
• Explainable diagnostics for normalisation and uncertain inferences.
• Visual references linking the JSON, the intermediate model, and the resulting code.
• Syntax highlighting, and download or copy of the generated files.
• Responsive interface, light and dark modes, and eight languages.
• Entirely local processing, privacy first.
• Static pages optimised for search engines.
• Automated tests across Chromium, Firefox, and WebKit.