AI-Copilot for Mobile App Navigation Implementation

We at TrueTech develop AI assistants for navigation in mobile applications. Complex mobile apps — banks, ERPs, medical platforms — lose users not because of missing features, but because finding the right feature is too difficult. The traditional answer — onboarding tours and help pages — works poorly: users go through the tour on first login and forget it a day later. An AI Copilot for navigation is an assistant that understands natural language requests and guides the user where they need to go. We guarantee that users will not get lost in the menu even in an app with 100+ screens.

What the Navigation Copilot Can Do

Not a 'chat', not an FAQ-bot. Three specific scenarios:

• Deep link navigation. User writes 'I want to transfer money to a card' — the assistant opens the correct screen. Technically: NLU model classifies the intent, maps it to a deep link, app performs navigation programmatically.

• Contextual hints. User lingers on a screen for three minutes without performing an action — Copilot offers help. Not a generic 'Need help?', but contextual: 'You are on the payment screen. Would you like me to explain the difference between transfer by phone number and by bank details?'

• Guided task execution. Multi-step tasks: 'apply for a mortgage' — 12 steps spread across three sections. Copilot guides step by step, tracks progress, and explains each screen.

How the AI-Copilot Understands User Intent

The most challenging part is mapping the user request to a specific action in the app. Two approaches:

• Classifier-based. Predefine a set of intents (50–200 for a typical app), train a classifier. Fast, predictable, cheap at runtime. Fails on non-standard phrasing.

• LLM + function calling. Describe all screens and actions as a set of functions. The LLM selects the appropriate function based on the user request:

// iOS — description of navigation functions for LLM
let navigationTools: [ChatCompletionTool] = [
    ChatCompletionTool(
        type: .function,
        function: ChatCompletionToolFunction(
            name: "navigate_to_screen",
            description: "Opens an app screen by its identifier",
            parameters: NavigationParameters.schema  // {screen_id: string, params: object}
        )
    ),
    ChatCompletionTool(
        type: .function,
        function: ChatCompletionToolFunction(
            name: "highlight_element",
            description: "Highlights a UI element on the current screen with an explanation",
            parameters: HighlightParameters.schema
        )
    ),
    ChatCompletionTool(
        type: .function,
        function: ChatCompletionToolFunction(
            name: "start_guided_flow",
            description: "Starts a step-by-step guide for a multi-step task",
            parameters: FlowParameters.schema
        )
    )
]

// Request with function calling
let request = ChatCompletionRequest(
    model: "gpt-4o-mini",
    messages: [systemMessage, userMessage],
    tools: navigationTools,
    toolChoice: .auto
)

The LLM returns tool_calls with function name and parameters; the app executes navigation.

How Guided Task Execution Works

The Copilot remembers the sequence of steps required to complete a task. It doesn't just open a screen; it guides the user through each step, checks completion, and returns to previous steps if needed. We describe in the system prompt the structure of each step: current screen, expected action, possible errors. For example, for a mortgage application: step 1 — select program, step 2 — upload documents, step 3 — confirm status. The Copilot tracks progress via session context.

Programmatic Navigation in iOS and Android

On iOS (SwiftUI) — via NavigationPath or custom Router:

class AppRouter: ObservableObject {
    @Published var path = NavigationPath()

    func navigate(to screen: AppScreen, params: [String: Any] = [:]) {
        switch screen {
        case .transfer:
            path.append(TransferRoute(params: params))
        case .loanApplication:
            path.append(LoanApplicationRoute(params: params))
        // ...
        }
    }

    // Called from AI Copilot
    func executeNavigationAction(_ action: NavigationAction) {
        DispatchQueue.main.async {
            self.navigate(to: action.screen, params: action.params)
        }
    }
}

On Android (Compose) — via NavController:

fun handleCopilotAction(action: NavigationAction, navController: NavController) {
    when (action.screenId) {
        "transfer" -> navController.navigate(
            "transfer?amount=${action.params["amount"] ?: ""}"
        )
        "loan_application" -> navController.navigate("loan/application")
        // ...
    }
}

UI Element Highlighting

Guided mode with element highlighting is technically more complex than navigation. You need an element identification system independent of screen position. On iOS: a tagging system via accessibilityIdentifier. The Copilot knows element names; an overlay layer draws a highlight with animation over the required element. On Android: similarly via contentDescription or custom tags + ViewTreeObserver to get element coordinates at runtime.

Contextual Awareness

The Copilot must know where the user is right now. Current screen, steps already completed, unfilled fields — this context is injected into the system prompt:

func buildCopilotContext(currentScreen: AppScreen, formState: FormState?) -> String {
    var context = "Current screen: \(currentScreen.name).\n"
    if let form = formState {
        context += "Completed fields: \(form.completedFields.joined(separator: ", ")).\n"
        context += "Missing required fields: \(form.missingRequired.joined(separator: ", ")).\n"
    }
    return context
}

Step-by-Step Implementation Guide for AI Copilot

1. Screen inventory — describe all screens and actions in the app, create an intent-to-deep-link mapping.
2. Select approach — we recommend LLM + function calling for flexibility.
3. Integrate with LLM — add API calls to the chosen model (GPT-4o, Claude, etc.).
4. Implement programmatic navigation — define routes via NavigationPath (iOS) or NavController (Android).
5. UI overlay — add a highlight layer for guided mode.
6. Testing — run an A/B test with task completion rate metrics.
7. Iteration — analyze logs and refine prompts.

Our Process

Screen and action inventory → intent schema design → NLU implementation (classifier or LLM function calling) → programmatic navigation system → UI overlay for element highlighting → guided flow engine → A/B test with task completion rate metrics.

Deliverables

• Intent schema and screen mapping
• LLM integration with function calling (Swift/Kotlin)
• Programmatic navigation for iOS and Android
• UI overlay container with element highlighting
• Guided flow engine
• Documentation for extension and maintenance
• Team training

Our Expertise and Experience

We are a team of certified Apple and Google engineers. 5+ years in mobile development, 30+ successful projects in FinTech, HealthTech, and Retail. We provide quality guarantees at every stage — each Copilot module undergoes load testing and code review.

Timeline Estimates

MVP with LLM function calling and basic navigation — 2–3 weeks. Full system with guided flows, element highlighting, and contextual hints — 3–5 weeks. Iterative refinement based on analytics — ongoing.

We will assess your project for free — contact us to discuss details. Get a consultation 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.