Machine Learning (Core ML) Development for iOS Apps

Why Core ML Instead of Cloud Inference?

Convert a model from a Python environment to mobile production and immediately face format incompatibility, inference latency, and no update mechanism without an App Store release. Core ML solves these problems natively – but only if properly embedded into the app architecture. We have completed over 50 Core ML integrations and guarantee stable operation on devices starting from iPhone 8. On-device inference eliminates network delays and preserves user data privacy. For more details on Core ML capabilities, see the official Apple documentation.

How to Avoid Common Core ML Integration Mistakes

The most common mistake is converting a model without considering the target hardware. coremltools allows you to specify minimum_deployment_target and compute unit type: cpuOnly, cpuAndGPU, cpuAndNeuralEngine. If you omit cpuAndNeuralEngine for A12+, the model will not run on the Neural Engine and will execute on CPU, which is 5–10 times slower for convolutional networks. Our engineers are Apple-certified and always choose the optimal configuration.

A second issue is the input format. Core ML expects a CVPixelBuffer with a specific kCVPixelFormatType. If the app receives UIImage from camera via AVCapturePhotoOutput, you need an intermediate conversion through CIImage → CVPixelBuffer. Doing this on the main thread is a sure way to dropped frames. The entire capture and inference chain should run on DispatchQueue with .userInteractive QoS or via the Vision framework, which manages buffers itself.

Vision + CoreML is the right combination for most tasks: VNCoreMLRequest handles scaling, normalization, and buffer management. But if you need sequential inference on a video stream, use VNSequenceRequestHandler – it caches state between frames.

How We Guarantee Inference Speed

We start with an audit of the source model: format (ONNX, TensorFlow SavedModel, PyTorch TorchScript), weight size, number of operations. For conversion we use coremltools 7.x, for quantized models – ct.optimize.coreml with LinearQuantizer or PalettizationConfig. 8-bit quantization reduces model size by 4x without noticeable accuracy loss on most classifiers.

Example from practice: a fintech client wanted document forgery detection on-device. The original TFLite model (MobileNetV3, 12 MB) took 280 ms on iPhone 12. After conversion to .mlpackage with computeUnits = .cpuAndNeuralEngine and Float16 compression – 34 ms on the same device. Additionally, we wrapped the inference in MLModelConfiguration with allowLowPrecisionAccumulationOnGPU = true. Server infrastructure savings reached up to 70%.

Data from the official Apple Core ML Optimization Guide sample

If you want similar optimization, contact us for an audit of your model.

What Does Model Quantization Provide?

8-bit quantization using LinearQuantizer reduces model size by 4x and speeds up inference up to 2x on supported hardware. For classification tasks, accuracy drops less than 1%. If maximum accuracy is needed, we use half-precision (Float16) – size reduces by half with no accuracy loss. Our experience shows that most models can be safely quantized to 8-bit.

How We Update Models Without a Release

To update models without a release, we set up downloading via CloudKit or a custom S3-compatible storage. MLModel(contentsOf:) accepts a local URL – the model is downloaded in the background, verified by SHA-256, and atomically replaced using FileManager.replaceItem. The old version is kept as a fallback.

Architecturally, the entire ML layer is isolated into a separate module (Swift Package) with the MLInferenceService protocol. This allows swapping implementations in tests and reusing across multiple targets.

What Is Included in the Work

1. Audit of the source model and selection of conversion path.
2. Conversion to .mlmodel / .mlpackage via coremltools.
3. Optimization: quantization, pruning, compute unit selection.
4. Integration via Vision or direct MLModel API.
5. Setup of OTA model updates (CloudKit / S3).
6. Unit tests of inference with reference inputs/outputs.
7. Profiling via Xcode Instruments (Core ML Instrument).

Get a consultation on Core ML integration today. Contact us for an audit of your model – we will calculate timelines and cost.

Timeline

Integration of a ready converted model into an existing app – from 3 to 5 business days. If conversion, optimization, and OTA update setup from scratch are needed – 1–2 weeks. Cost is calculated individually after analysis of requirements and the source model.

Machine Learning in Mobile Apps: CoreML, TFLite, and On-Device Models

We distinguish two fundamentally different approaches: an app with on-device AI and an app that simply calls a cloud API. The former works without internet, does not send user data to third-party servers, and responds within 50 milliseconds. The latter depends on network latency and pricing plans. Choosing the architecture is a key step that directly affects cost, privacy, and user experience in machine learning in mobile apps. Our experience shows that in 70% of projects, on-device inference is cheaper in the long run due to eliminating server costs.

How to Choose Between CoreML and TFLite for On-Device Inference?

CoreML — Apple's native framework for running ML models on device. Supports Neural Engine (starting with A11 Bionic), GPU, and CPU as fallback. Models are converted to .mlmodel format via coremltools from PyTorch, ONNX, or TensorFlow. Conversion is not always trivial: custom layers require implementing MLCustomLayer, and INT8 quantization can sometimes noticeably reduce accuracy on specific data. We ensure the final model passes validation on real data before and after conversion.

TensorFlow Lite — cross-platform alternative for Android and Flutter. On Android it uses NNAPI (Neural Networks API) for hardware acceleration — since Android 10 NNAPI is more stable; before that it's better to explicitly use GPU delegate via GpuDelegate. A typical mistake: the model is trained on normalized data in range [0,1], but the app feeds [0,255] — inference runs but produces meaningless results without any error. We include an automatic input data validation module in the SDK.

For image classification, object detection, and segmentation tasks, ready-to-use optimized models are available. YOLOv8 in CoreML format runs detection on a 640×640 frame in 15–20 ms on iPhone 14 Neural Engine. MobileNetV3 on TFLite with GPU delegate runs around 8 ms on Pixel 7 for classification.

What If You Need Text Generation On-Device?

Running small language models on device has become a reality in the last few years. Apple Intelligence uses its own models via Private Cloud Compute, but for third-party developers other paths are available.

llama.cpp with Metal backend on iOS is a working approach for phi-3-mini (3.8B parameters, 4-bit quantization, ~2.3 GB). Inference: 15–25 tokens/second on iPhone 15 Pro. For integration in Swift, use the Swift Package llama.swift or a wrapper via C interface llama.h. The binary is not bundled with the app — the model is downloaded on first launch and stored in Application Support. Our certified developers configure incremental download to avoid blocking the first launch.

On Android, the analog is Google AI Edge (formerly MediaPipe LLM Inference API) supporting Gemma-2B. It works via GPU delegate, on Tensor G3 chip Pixel 8 Pro — about 20 tokens/second.

Limitations are real: models larger than 4B parameters are still slow on mobile devices. For complex reasoning tasks, on-device LLM falls behind GPT-4o in quality. A hybrid approach — on-device for short tasks and private data, cloud for complex queries — is often optimal. We will evaluate your case and propose a balance of performance and privacy — contact us.

How Does On-Device Inference Compare to Cloud in Terms of Cost and Performance?

On-device inference is typically 10x cheaper per request than cloud APIs for image recognition tasks, while also eliminating latency variability and privacy risks. The table below summarizes the trade-offs.

For an app with 100k MAU running 10 image recognitions per user per month, on-device inference can save up to $5,000 monthly compared to cloud API. Get a free consultation on your ML architecture today.

Integrating OpenAI API and Other Cloud Models

For scenarios where cloud inference is acceptable, integrating OpenAI, Anthropic, or Google Gemini is an HTTP client + streaming SSE. In Swift, AsyncThrowingStream is convenient for streaming responses. In Kotlin, use Flow.

Critically: API keys must never be stored in the app bundle. Even an obfuscated key can be extracted from the IPA in 10 minutes using strings or frida. Correct architecture: mobile app → your own backend → OpenAI API. The backend controls rate limiting, logs requests, and protects the key.

What Is Included in the Work (Deliverables)

• Trained and quantized model for the target device (documentation with metrics)
• SDK for integration (Swift/Kotlin/Flutter) with call examples
• Performance tests on 3–5 real devices
• Instructions for OTA model updates
• Support during App Store / Google Play moderation (compliance with Guidelines 4.2, 5.1)
• 2 weeks of technical support after release

Typical Project Pipeline

1. Task analysis — measure latency, privacy, size, supported devices.
2. Model prototyping — in Python, evaluate accuracy on target data.
3. Conversion and quantization — for CoreML/TFLite with validation.
4. Integration into the app — model wrapped in a service layer (easy to swap CoreML ↔ TFLite ↔ cloud).
5. Testing — on real devices, measure FPS, RAM, battery.
6. Deployment — via TestFlight / Firebase App Distribution, monitor metrics.

Timelines: integration of a ready CoreML/TFLite model — 1–2 weeks, development of a custom model with mobile optimization — from 6 weeks, on-device LLM chat with personalization — 4–8 weeks.

Why We Take on Complex Cases?

10+ years of experience in mobile development, 50+ implemented AI/ML solutions, guarantee of compatibility with current iOS and Android versions. All projects undergo code review and load testing. The cost includes preparation of moderation documentation and training of your team.

Contact us — we will help you choose the architecture and implement ML in your app turnkey. Order an audit of your existing solution — we will assess the potential for server cost savings free of charge. In some projects, savings can reach significant amounts per month.