Smart Reply for iOS and Android: From ML Kit to Custom LLMs

How Smart Reply Solves the Speed Problem in Chat Communication?

We integrate Smart Reply into messengers, CRMs, and e-commerce platforms, and we see how this feature dramatically accelerates conversations. The user gets three ready-made reply options below each message — one tap is enough. Without Smart Reply, they must manually type "Okay", "Got it", "Thanks". This slows down the dialogue and increases the number of unfinished messages. Research by Google ML Kit shows that the feature boosts engagement and sent message volume by 15–30%. For businesses, this means faster query processing and higher customer loyalty. In our practice, clients save 2–3 seconds per reply, which accumulates to hours saved per day for the entire support team. In one project for an online store, implementing Smart Reply reduced the average operator response time from 45 to 15 seconds, and the number of messages per dialogue increased by 20%. If you are evaluating Smart Reply implementation, contact us for a preliminary assessment.

Why ML Kit Smart Reply Doesn't Fit Russian Language?

Google's ready-made SmartReply model works only with English. For Android, integration takes an hour:

val smartReply = SmartReply.getClient()

val conversation = messages.takeLast(10).map { msg ->
    if (msg.isFromUser) {
        TextMessage.createForLocalUser(msg.text, msg.timestamp)
    } else {
        TextMessage.createForRemoteUser(msg.text, msg.timestamp, msg.senderId)
    }
}

smartReply.suggestReplies(conversation)
    .addOnSuccessListener { result ->
        if (result.status == SmartReplySuggestionResult.STATUS_SUCCESS) {
            val suggestions = result.suggestions.map { it.text }
            showSuggestions(suggestions)
        }
    }
    .addOnFailureListener { /* hide UI */ }

Plus side: speed <20 ms. Minus: limited template set and only English. For iOS, the alternative via Natural Language or Apple Intelligence API exists but offers poorer functionality. If your app targets Russian-speaking audience, ML Kit is not suitable — a custom model is needed. Also consider App Store Review Guidelines: section 5.1 permits on-device ML without special permission, but custom LLMs processing user data must comply with privacy rules.

How to Implement Contextual Smart Reply on a Custom LLM?

For Russian and specific domains (support, healthcare, B2B), we use an LLM with a prompt. Example in Swift:

func generateReplySuggestions(
    lastMessages: [ChatMessage],
    count: Int = 3
) async -> [String] {
    let context = lastMessages.suffix(5)
        .map { "\($0.role): \($0.text)" }
        .joined(separator: "\n")

    let prompt = """
    You help the user quickly reply to a chat message.
    Dialogue history:
    \(context)

    Suggest \(count) short reply options for the user.
    Each reply is one sentence, maximum 10 words.
    Format: JSON array of strings.
    """

    let response = try await llmClient.complete(prompt: prompt, maxTokens: 100)
    return parseJSONArray(response) ?? []
}

A similar implementation on Android with Kotlin and ML Kit or a custom model:

suspend fun generateReplySuggestions(
    lastMessages: List<ChatMessage>,
    count: Int = 3
): List<String> {
    val context = lastMessages.takeLast(5)
        .joinToString("\n") { "${it.role}: ${it.text}" }

    val prompt = """
    You help the user quickly reply to a chat message.
    Dialogue history:
    $context

    Suggest $count short reply options for the user.
    Each reply is one sentence, maximum 10 words.
    Format: JSON array of strings.
    """

    val response = llmClient.complete(prompt, maxTokens = 100)
    return parseJSONArray(response) ?: emptyList()
}

Latency of 1–2 seconds is acceptable. We preload options while the user reads — by the time they're ready to reply, suggestions are ready. Custom LLM is 50x slower than ML Kit but provides Russian language and context flexibility. On-device models (TensorFlow Lite, Core ML) are faster but require more memory and are less configurable.

What to Choose: ML Kit or Custom LLM?

For English and simple scenarios, ML Kit is better (50x faster). For Russian and specific needs, custom model.

How Does Smart Reply Differ on iOS and Android?

When to Show Smart Reply?

Smart Reply appears after an incoming message and disappears when the user starts typing. Three suggestions is optimal (Google Research). More overloads, less gives no choice. We use chips (horizontal scroll): MaterialChip on Android, custom Chip in SwiftUI.

// Android: hide when typing
editText.addTextChangedListener(object : TextWatcher {
    override fun onTextChanged(s: CharSequence?, start: Int, before: Int, count: Int) {
        smartReplyChips.isVisible = s.isNullOrEmpty()
    }
    override fun afterTextChanged(s: Editable?) {}
    override fun beforeTextChanged(s: CharSequence?, start: Int, count: Int, after: Int) {}
})

Also important to set up analytics: track how many users use suggested replies, and A/B test the number of chips.

How Does the Implementation Process Work?

We work in stages:

1. Analyze Smart Reply usage scenarios in your app.
2. Choose approach: ML Kit or custom LLM.
3. Design architecture (preloading, caching).
4. Implement on Android (Kotlin/Compose) and iOS (Swift/SwiftUI).
5. Integrate with chat and analytics system.
6. Code documentation and instructions for your team.
7. Test on real dialogues.
8. Post-implementation support: bug fixes and refinements.

Deliverables include:

• Integration and setup documentation.
• Access to test environment for validation.
• Team training (1–2 hours) on using the solution.
• One month of technical support after launch.

What Are the Timelines and Results?

Smart Reply via ML Kit (Android, English) — 1–2 days. Custom on LLM with context classification — 5–8 days. Full integration on both platforms — up to 2 weeks. Based on our projects, implementation increases engagement by 20–40% and reduces average response time by 2–3 times. For a support team of 10 people, time savings amount to up to 500 person-hours per month, converting to financial savings of $3,000 to $10,000 monthly. Contact us to discuss your app's needs.

Our Experience

We have implemented Smart Reply for messengers, CRMs, and e-commerce. We work with ML Kit and custom models. Five years in the market, over 50 projects. We guarantee correct operation on Android and iOS. Request a consultation — we'll evaluate your scenario and propose the optimal solution. Get demo access to a working example today.

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.