AI-Powered Automatic Content Tagging in Mobile Apps

A typical scenario: a user uploads a photo to an app, but to add a tag, they must manually choose from hundreds of options or type text. Context and time are lost. We implement automatic AI-based tagging — images, text, and video get labels without human intervention. Our track record: over 20 projects, from marketplaces to social networks, with custom taxonomies. The solution applies to any domain: medicine, real estate, retail, education. On-device models achieve 92% accuracy, server models 98% with a properly tuned confidence score. We guarantee 98% accuracy threshold adjustment and have over 10 years of experience in AI and mobile development.

For example, an online clothing store: we trained a model to recognize 150 categories with 91% accuracy. Now every uploaded item automatically receives tags like “Dress”, “Cotton”, “Summer”. This cut moderation time by 4x and saved roughly 500,000 rubles per year in manual labeling. Our clients typically save $10,000–$50,000 per year depending on content volume.

Apple Core ML documentation recommends transfer learning for creating compact models with 20+ examples per category.

How AI Tags Images in iOS and Android

The standard stack is VNClassifyImageRequest (iOS) + ImageLabeler (Android). They return generic labels like “Food”, “Sky”, “Cat”. For business needs, you need a custom taxonomy: not “Clothing”, but “Leather jacket”, “Floral dress”. We train a custom model using CreateML (iOS) or TensorFlow Lite (Android). Below is an example of training a custom model under iOS.

// Training via CreateML (run on Mac, not on device)
import CreateML

let trainingData = MLImageClassifier.DataSource.labeledDirectories(
    at: URL(fileURLWithPath: "/training_data")
    // Structure: /training_data/jacket/, /training_data/shoes/, /training_data/bag/
)

var params = MLImageClassifier.ModelParameters()
params.maxIterations = 25
params.validationData = .split(strategy: .automatic)
params.featureExtractor = .scenePrint(revision: 2)  // Transfer learning from Apple

let model = try MLImageClassifier(trainingData: trainingData, parameters: params)
try model.write(to: URL(fileURLWithPath: "/model.mlmodel"), metadata: nil)

20–50 examples per category, 15–30 minutes of training on a MacBook Pro M2 — you get a compact model. Core ML Model Deployment allows updating it without publishing a new App Store version.

Why Hierarchical Tags Speed Up Search

A flat list of tags is chaos. A hierarchy like “Food → Italian cuisine → Pasta” gives structured search and filters. We implement this via trees:

// Android: TagTree
data class Tag(
    val id: String,
    val name: String,
    val parentId: String?,
    val synonyms: List<String> = emptyList()
)

// When tagging: if tag "Pasta" is assigned, automatically add parent tags
fun expandWithParents(tagId: String, tagTree: Map<String, Tag>): Set<String> {
    val result = mutableSetOf(tagId)
    var current = tagTree[tagId]
    while (current?.parentId != null) {
        current = tagTree[current.parentId]
        current?.let { result.add(it.id) }
    }
    return result
}

For storage we use a separate table with a source field (auto, user, admin). Auto-tags are visible only in search, user tags in the UI.

On-Device vs Server Tagging Comparison

We combine both: basic tags are set on the device, for complex cases we send a request to the server. On-device tagging is 3–10x faster than server with similar accuracy for typical categories.

How to Improve Tag Accuracy

If accuracy is below expectations, increase the training set to 100+ examples per category or use a server-side model for difficult cases. Regular retraining on new data keeps the taxonomy up-to-date. We recommend retraining the model monthly as new content types appear.

Tagging Text and Video

Text posts — NLP classification on-device via the Natural Language Framework or on the server. Prompt: "Determine 3–5 tags from the list: ...". JSON response is parsed on the client.

Video — key frame analysis:

func tagVideo(at url: URL) async throws -> Set<String> {
    let asset = AVURLAsset(url: url)
    let duration = asset.duration.seconds
    let generator = AVAssetImageGenerator(asset: asset)
    generator.maximumSize = CGSize(width: 224, height: 224)

    var allTags = Set<String>()
    var time = 0.0
    while time < duration {
        let cgImage = try generator.copyCGImage(at: CMTime(seconds: time, preferredTimescale: 600), actualTime: nil)
        let frameTags = try await classifyImage(cgImage)
        allTags.formUnion(frameTags)
        time += 3.0  // every 3 seconds
    }
    return allTags
}

For long videos we use background tasks (BackgroundFetch on iOS, WorkManager on Android) or send to the backend.

Implementation Steps

1. Content Audit & Requirements: Analyze current content volume and types, define business goals.
2. Taxonomy Design: Create flat or hierarchical tag structure with stakeholder input.
3. Data Collection & Annotation: Gather minimum 20 examples per category, annotate manually.
4. Model Training: Use CreateML (iOS) or TensorFlow Lite (Android) with transfer learning; validate accuracy.
5. Integration: Embed SDK into app, connect on-device and server models, implement search filters.
6. Testing & Deployment: A/B test thresholds, deploy to production, monitor performance.

Deliverables

• Audit report and taxonomy design document
• Annotated training dataset
• Trained model (.mlmodel / .tflite) and source code
• Integrated iOS (Swift) and Android (Kotlin) SDK with hierarchy support
• On-device + server architecture (Firebase, Supabase, or your backend)
• Accuracy testing results and threshold recommendations
• Complete API and taxonomy documentation
• Team training (2 workshops)
• One month post-release technical support with access to source code and model artifacts

Implementation Timeframes

Pricing is determined individually. For an accurate estimate, send us your project description — we will prepare a commercial proposal within 1–2 days.

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.