Learning AI as a Mobile Developer — Why I'm Exploring Local LLMs for Mobile Apps
Journey Context
After 6.5 years of building react-native apps. Most of my work has been focused on building production apps, integrating APIs, optimizing performance, and shipping features. But instead of just calling another API, I want to see if we can make the phone think for itself.
But recently, one thing has become very clear.
AI is becoming part of almost every product.
Instead of ignoring it or treating it as a “backend/data science thing”, I decided to understand how AI can actually be used inside mobile apps.
Not from a research perspective. Not from a deep ML theory perspective.
But from a developer perspective.
My goal is simple:
Learn how mobile developers can integrate AI into apps in practical ways.
And that led me to an interesting area:
Local LLMs running directly on mobile devices.
This blog marks the start of my learning journey.
Developer Perspective
In React Native, we usually think about fetch() calls. With Local LLMs, the "backend" is just another local library. Imagine:
An app that summarizes your private notes without ever sending them to a server.
A chat assistant that works in "Airplane Mode."
Zero monthly costs for AI features, no matter how many users you have.
Struggles / Confusions
I’m still a bit confused about Model Sizes. I see models labeled "1B", "3B", or "7B". Apparently, "B" stands for billions of parameters. As a dev used to keeping app bundles under 50MB, the idea of a 2GB "lightweight" model is a bit terrifying. How do we ship that to the App Store without users hating us?
What Surprised Me
I was shocked to find out that modern mobile chips (like Apple’s M-series or the latest Snapdragon) have dedicated "Neural Engines." Our users are literally walking around with AI hardware that most apps aren't even using yet.
Earlier LLMs were massive.
But now we have quantized models that can run in just a few GB of memory.
That means modern phones can actually handle them.
Of course, there are still challenges:
model size memory usage
battery impact performance on low-end devices
But it's becoming practical.
Closing Thoughts
I’m not approaching this as an AI expert.
I’m approaching this as a mobile developer trying to understand how AI fits into modern apps.
The goal is simple:
Learn AI by building things.
And document everything along the way.
If you're also a mobile developer curious about AI, this journey might help you too.
More experiments coming soon.
