Artificial intelligence is no longer reserved for tech giants. With Core ML, Create ML, and cloud APIs, any mobile developer can enrich their apps with AI features. Here's how to proceed pragmatically.
Core ML: On-Device AI
Core ML allows running machine learning models directly on the device, without internet connection. It's ideal for data privacy and app responsiveness.
Apple provides pre-trained models for vision (object detection, image classification), natural language processing, and sound recognition. These models are optimized for Apple Silicon chips.
For specific use cases, Create ML allows training your own models directly on Mac, without deep ML knowledge.
import CoreML
import Vision
func classifyImage(_ image: UIImage) async throws -> String {
guard let model = try? VNCoreMLModel(for: MobileNetV2().model) else {
throw ClassificationError.modelLoadFailed
}
let request = VNCoreMLRequest(model: model)
let handler = VNImageRequestHandler(cgImage: image.cgImage!)
try handler.perform([request])
guard let results = request.results as? [VNClassificationObservation],
let topResult = results.first else {
throw ClassificationError.noResults
}
return topResult.identifier
}Cloud APIs: OpenAI, Claude, Gemini
For more complex tasks like text generation or advanced semantic analysis, cloud APIs are essential. OpenAI GPT-4, Anthropic Claude, and Google Gemini offer impressive capabilities.
The recommended architecture goes through a backend (Edge Function, API Gateway) that manages API keys and rate limiting. Never store API keys in client code.
Implement smart caching to reduce costs and improve latency. Similar responses can be cached for an appropriate duration.
Concrete Use Cases
Automatic photo classification: use Core ML to automatically sort photos by category (landscapes, people, documents).
Smart suggestions: analyze user behavior to offer personalized recommendations. Core ML can run lightweight recommendation models.
Voice transcription: Apple's Speech Recognition offers accurate and free transcription, directly on-device since iOS 17.
Contextual chatbot: integrate an API like Claude for a conversational assistant that understands your app's context.
Best Practices
Always prioritize on-device processing when possible. It's faster, less expensive, and better respects privacy.
Plan for fallbacks: if the ML model fails or the cloud API is unavailable, the app should remain functional.
Test your models with diverse data. Biases in training data can create unexpected behaviors.
Measure the impact on battery and performance. ML models can be resource-intensive.
Conclusion
Integrating AI into mobile apps is accessible to all developers. Start with simple use cases with Core ML, then explore cloud APIs for more advanced features. The important thing is to keep the user at the center: AI should improve the experience, not complicate it.