IDU

Inspiration

The inspiration came from a simple observation: the "vibe-coding" revolution is booming on desktops, but mobile devices have been completely left behind. We watched developers and non-technical creators ship incredible projects using natural language at their desks, yet found themselves powerless when an idea struck on the subway, at a cafe, or during a walk. We wanted to eliminate the desktop boundary and turn the smartphone into a fully capable, prompt-driven software factory. If you can explain your idea to a friend via text, you should be able to build it on your phone.

What it does

Our project is a mobile-first AI IDE designed specifically for rapid vibe-coding. Users open the app and use voice or text prompts to describe any micro-app, utility, tracker, or interactive widget they need on the spot. The AI instantly generates the UI, logic, and state management, running the functional application inside a secure mobile sandbox. Users can test the app immediately, fine-tune the design or logic using follow-up prompts ("make it dark mode," "add a charts view"), and instantly publish it via a QR code or web link for anyone else to use.

How we built it

• Mobile Shell: Built with React Native and Expo to ensure a fluid, native feel across both iOS and Android.
• Dynamic Rendering Engine: Developed a custom interpreter that translates structured JSON and lightweight JavaScript code generated by the LLM into native, reactive UI components on the fly.
• AI Orchestration: Integrated Anthropic's Claude 3.5 Sonnet and OpenAI's GPT-4o APIs with highly optimized system prompts designed to prevent code syntax errors and enforce mobile responsiveness.
• Backend-as-a-Service: Utilized Supabase for rapid user authentication, real-time database syncing, and hosting the generated micro-apps.

Challenges we ran into

• Latency vs. Vibe: Waiting 30 seconds for an LLM to generate code kills the "vibe." We solved this by streaming the JSON structure and rendering UI components block-by-block in real-time as the AI thought, keeping the user engaged.
• Mobile Screen Constraints: Traditional code editors don't fit on a 6-inch screen. We had to completely reinvent the "IDE layout," replacing code trees and text files with a visual layer stack and a chat-centric iteration interface.
• Sanitization & Security: Executing dynamic code on a mobile device poses security risks. We had to build a robust validation layer to sanitize the AI's output and sandbox the execution environment.

Accomplishments that we're proud of

• True Zero-to-App in 30 Seconds: We successfully built a pipeline where a spoken prompt generates a fully interactive, stateful mobile widget with a working database in under half a minute.
• Elegant Visual Context: Our UI doesn't just show text; it provides a beautiful visual layer inspector that lets users see exactly what components the AI generated and touch them to modify them.
• Flawless Voice-to-Code: The voice prompt integration feels entirely natural, allowing users to build software purely by talking to their phone.

What we learned

• Prompt Engineering is the New Compiler: We learned that context window management and strict system prompting are just as critical as traditional memory management when it comes to runtime stability.
• Constraints Breed Creativity: Designing a development environment for a touch-first, small-screen device forced us to strip away the bloat of traditional IDEs and focus entirely on user intent and instant feedback loops.

What's next for IDU

• Offline Vibe-Coding: Integrating lightweight, on-device LLMs so users can prompt and build tools even when they have no cellular service or internet connection.
• Native Feature Access: Expanding the AI's capability to safely tap into device hardware, such as the camera, GPS, accelerometer, and local push notifications.
• The Vibe Marketplace: A community feed within the app where creators can publish their vibe-coded micro-apps, allowing other users to remix, fork, and build on top of them instantly.

Built With

• css
• react