Show IH: I ported my Android camera app to iOS in 11 days using Google AI Studio. Here's where I had to ignore the AI.

Hey hackers,

Porting a native camera app from Android to iOS is usually a nightmare. You’re translating entirely different APIs and rendering pipelines. As a solo dev running Muxixi Studios, I didn't have months to spare.

To speed up the iOS port of TimerCamera+, I ran an experiment using Gemini 3.1 Pro Preview as my senior pair programmer. The result? Zero codebase to a live App Store MVP (v1.1) in exactly 11 days.

Here is what the workflow looked like, the tools I used, and where I had to push back against the AI:

1. The Secret Weapon: Repomix + 1M Token Context
How do you feed an entire Android project to an AI without it losing its mind? Even though Gemini 3.1 Pro Preview has a massive 1 Million token context window, dumping raw folders into a prompt is a recipe for disaster.

I highly recommend a tool called Repomix. It allows you to pack specifically targeted directories (in my case, my Android CameraX logic) into a single, clean, AI-readable file. Crucially, Repomix gives you a clear sense of the codebase's token magnitude before you hit send. You know exactly how much of that 1M context window you are eating up.

I passed this perfectly structured context to Gemini and prompted it to map the architecture to Apple's native APIs. It provided a remarkably solid foundation almost instantly.

2. The SwiftUI vs. Metal Debate (Ignoring the AI)
Here is where the model and I had a fundamental disagreement. Gemini strongly pushed me to rewrite the entire camera preview and UI layer purely in SwiftUI.

I refused. Instead, I insisted on building a Metal + SwiftUI hybrid architecture.

I used native Metal exclusively for the core rendering pipeline to keep it lightweight, and restricted SwiftUI purely to the main layout and interactions. I used Gemini to generate the boilerplate Metal shaders, then manually refined the integration.

Why I'm glad I stuck to my guns:
This hybrid decision was the game-changer. It solved the performance bottlenecks immediately while keeping UI development blazing fast. More importantly, it allowed me to make the UX and interactions almost 100% identical between the new iOS version and my mature Android version (currently v1.5.6).

3. The Result: A 3.7MB Powerhouse
Because I didn't rely on heavy third-party engines or bloated pure-UI rendering, the final iOS app size is a ridiculously small 3.7MB. It’s a hands-free selfie timer with zero ads and zero data collection.

AI is incredible for accelerating cross-platform ports, but this 11-day sprint taught me two things: a tool like Repomix is essential for managing context, and a developer's architectural intuition is still the most valuable asset.

Has anyone else here tried Repomix for feeding code to LLMs? How do you manage your context windows?

🍎 If you want to see how the Metal/SwiftUI hybrid turned out, here’s the iOS link: https://apps.apple.com/app/id6768857092

P.S. If you have any suggestions or feedback for TimerCamera+, please send them my way. It is incredibly important to me and helps shape the future of the app!