AI-Driven Zero-Compile Runtime Native App Engine
An experimental dynamic canvas framework that compiles natural language prompts into fully functional native applications at runtime, bypassing traditional compilation steps. Engineered complete with an on-the-fly local database, reactive state management, runtime scripting execution, dynamic native layouts, and an integrated AI architect capable of designing, building, and running the entire system interactively.
The Overview
Project 2360, internally codenamed Flowserract, is an experimental architectural framework designed to push the boundaries of how AI interacts with application state. It acts as a generative native UI engine that parses natural language prompts to dynamically assemble and render Flutter widgets in real-time.
The Engineering Challenge
Current AI applications usually just return text or simple HTML. Flowserract aims to return fully functional, compiled-feeling native components. The challenge lies in translating unstructured AI intent into the strict, typed widget tree required by Flutter, without causing layout shifts, jank, or application crashes.
The Implementation
By heavily leveraging the llm_json_stream parser, Flowserract listens to raw LLM token streams and instantly compiles them into structural JSON representations of native UI elements. As the AI "thinks" of a layout, the engine is already painting the buttons, containers, and text fields onto the canvas. It is an exploration in bridging complex AI logic directly to low-level component rendering, maintaining strict 60fps animations while the UI literally builds itself.