This is a guest post from Said Kaban - he is the CTO and Co-Founder of FERASET.
…
In late 2022, most mobile engineers I knew thought Expo was a prototyping tool. Something you'd use for a demo or a hackathon, then switch to the React Native CLI when you wanted to build anything real. Especially in Turkey, where we're based, this was the default take.
We thought differently. We looked at where Expo was going and saw the same trajectory as Next.js in the React ecosystem. An opinionated, managed layer that doesn't limit you but makes you way faster. So we went all in.
Today FERASET is an AI consumer app studio of ~40 people in Istanbul. We've hit #1 globally in the App Store in multiple countries including the US. Our entire mobile stack runs on Expo. It has since day one.
Here's how that played out.
We started at the right time, with the right bet
FERASET started right when the AI boom hit. ChatGPT had just launched. DALL-E was making headlines. Midjourney was everywhere. But they were all web products. If you wanted to create with AI, you were on a desktop or in a mobile browser. Nobody had built great native mobile experiences for AI yet and we knew that's where the users actually were.
So speed was the only thing that mattered. The teams that got polished AI apps into the stores first would win. Every week of delay meant losing ground.
We needed a framework that would let a small team ship to both platforms fast and not waste time on native toolchain issues. Expo gave us that. Music AI, our first app, went from the first line of code to live in the App Store in 4 weeks. A full MVP, both platforms, submitted and approved. That's the kind of speed Expo made possible and it's what let us move before the market got crowded.
Our first app hit #1 globally
Our first app was Music AI, an AI cover generation app. You could create AI covers of songs with different voices, or even train a custom voice of yourself or your loved ones and make them sing. It took off fast.
Music AI became the best AI cover app in the world. We surpassed Spotify in the App Store charts and reached the #1 spot globally in multiple countries including the US and major European markets.
That confirmed two things for us. The mobile AI market was real and massive. And Expo could absolutely handle production at scale. This wasn't a prototyping tool. This was the framework we'd build the whole company on.
From there, the next step felt obvious. The AI consumer space isn't a single-app market. Different users want different experiences: image generation, video, specific creative tools. One app can't serve all of them well. So instead of trying to stuff everything into Music AI, we decided to run FERASET as a studio. Multiple apps, each focused on a specific use case, all built by the same core team on Expo.
A real example of what we build on Expo
The best way to show what building on Expo actually looks like for us is through a specific example: expo-ai-kit.
expo-ai-kit: on-device AI for Expo apps
All of our apps use language models for things like prompt refinement, content guideline checks and text generation. Tasks that need an LLM but don't need a state-of-the-art model. For a long time, that meant API calls to cloud providers for every single request. It worked, but we were paying for inference on tasks that a small on-device model could handle just fine and user prompts were leaving the device when they didn't need to.
Apple Foundation Models on iOS and Google ML Kit's Prompt API on Android changed that equation. For the kind of tasks we need, these on-device models are more than good enough. And they're free. No API keys, no per-token costs, no network dependency. The problem was there was no clean way to use them from Expo. So we built expo-ai-kit.
It's an open-source Expo module that gives you a simple JavaScript API for on-device language models on both platforms. Here's what it looks like:
import { isAvailable, sendMessage } from 'expo-ai-kit';const available = await isAvailable();if (available) {const response = await sendMessage([{ role: 'user', content: 'Refine this prompt for image generation: a cat sitting' }],{ systemPrompt: 'You are a prompt engineering assistant.' });console.log(response.text);}
There's also streaming support for progressive text display:
import { streamMessage } from 'expo-ai-kit';const { promise, stop } = streamMessage([{ role: 'user', content: 'Check this text against content guidelines...' }],(event) => {setResponseText(event.accumulatedText);},{ systemPrompt: 'You are a content moderation assistant.' });
Just npm install expo-ai-kit, add the config plugin and you're running on-device inference. The library currently gets tens of thousands of weekly downloads on npm.
For us, tasks that used to cost money on every request now run for free on the device. Response times for simple tasks dropped because there's no network round trip. And as on-device models keep getting more capable with every new chip generation, the range of tasks we can offload from the cloud will only grow.
We open-sourced it because we figured other developers would run into the same problem. And building it reinforced something we'd been experiencing all along: expo-ai-kit is real Swift and Kotlin code, talking to platform-level AI frameworks, integrated through Expo's module system and distributed as a config plugin. No ejecting, no bare workflow. You can go deep into native when you need to and still stay in the managed ecosystem.
The hardest part wasn't the tech
The biggest challenge we faced was the developer ecosystem in Turkey.
When we started, most mobile engineers here were more comfortable with other frameworks. Expo wasn't really on their radar for production apps and to be fair, that was a reasonable take at the time given how young the ecosystem was. Getting experienced engineers to give it a serious look took some convincing.
Hiring was tough early on. We couldn't just post a listing for an Expo engineer and get a pile of qualified candidates. We had to find engineers who were open to it, educate them on why Expo was the right choice and prove it through results.
We spent time building up the React Native and Expo community in Turkey. Sharing what we were learning, making the case that the React ecosystem would win mobile the same way React won the web. We wanted to spearhead that in Turkey rather than wait for the market to catch up.
It worked. But it took effort that had nothing to do with writing code.
What surprised us about Expo
The stability of Expo's managed workflow at scale was the biggest surprise. We're on CNG which helps immensely because we ship so fast and we create different builds constantly: dev, preview, prod. The ease of creating those builds and distributing them to our product teams for testing and submission has been amazingly smooth.
When you're a small team, managed workflows feel like training wheels. You assume you'll outgrow them. We haven't. The team is bigger, the apps are more complex, the user base is in the millions and we're still fully managed. The benefits scale with your app and your team.
The upgrade path between SDK versions has also been smoother than expected. Upgrading a single app is one thing. Upgrading multiple apps that share code is a coordination challenge. Expo's upgrade tooling and migration guides make this significantly less painful than it could be.
And honestly, the pace of innovation from the Expo team keeps impressing us. They've consistently been ahead of the curve. File-based routing was a great example of that. They saw where the developer experience was going and got there early.
What's next?
Now that the era of agentic coding has officially taken over, Expo is pushing that front too. We're eagerly expecting the official launch of Expo Agent, which we believe will continue to power and accelerate our client-side development process. The idea of AI-assisted development built directly into the framework we already use is exactly the kind of forward-thinking move that made us bet on Expo in the first place.
On our side, we're continuing to expand our product portfolio. The AI consumer space is still growing fast and we see big opportunities in video generation and real-time AI experiences on mobile. All of it will be built on Expo.
If you're a small team somewhere thinking about building consumer apps, here's what I'd say: don't overthink your stack. Pick Expo, start shipping and focus your energy on the product. The framework will scale with you. Ours did, from a tiny team in Istanbul to apps that hit number one globally. And we're just getting started.
FERASET is an AI consumer app studio based in Istanbul. Our apps are available on the App Store and Google Play. As is standard practice among AI app studios and mobile gaming studios, our apps are published under different developer accounts for various operational purposes, so you won't find them all under the FERASET name.
