Bot Assistant Development for Mobile Apps with State Machine

Why a bot assistant is more than just a chat?

We develop conversational interfaces for mobile apps that automate order intake. This is not a chat with an operator: every action — create, modify, cancel — is atomic and rollbackable. The user can change their mind at any step, and the bot retains context.

In our practice, we have completed over 50 projects integrating bots with OMS. For example, for a coffee chain we built a bot that handles 300+ orders per day. Average checkout time is 30 seconds — twice as fast as linear scripts. Such solutions reduce support load by 40% and offer significant cost savings over time. Contact us to assess the potential for your case.

How state machine prevents data loss?

Completing an order via bot is a multi-step form stretched over time. Between replies, the user might close the app, switch to another, and return after 10 minutes. The state must be preserved. We use a state machine on Redis: each step is a finite automaton with explicit transitions. Wikipedia - Finite-state machine

The slot structure for an order dialogue:

{
  "session_id": "uuid",
  "step": "confirm_address",
  "order_draft": {
    "items": [
      {"sku": "ITEM-123", "qty": 2, "price": 1500}
    ],
    "delivery_address": null,
    "payment_method": "card",
    "promo_code": null
  },
  "expires_at": "2025-01-15T14:30:00Z"
}

State is stored on the server (Redis with TTL 30–60 minutes). The mobile app sends only session_id with each message.

A critical point: every dialogue step must support 'back' and 'cancel' commands. If the user writes 'change item' at the address confirmation step, the bot should return to the item selection step without wiping already entered data.

State machines yield 2–3x fewer bugs than custom dialogue handlers — proven across dozens of projects. A state-machine bot processes orders twice as fast as conventional scripts.

How we integrate the bot with the order backend?

The bot should not contain business logic. It calls API methods:

• POST /orders/draft — create a draft
• PUT /orders/draft/{id}/items — modify items
• POST /orders/draft/{id}/submit — finalize
• DELETE /orders/{id} — cancel

A typical issue: the bot lets the user add an out-of-stock item, which is only discovered at final submit. That's poor UX. Stock checks must happen when adding an item to the draft.

If an LLM is used for message processing, function calling makes the integration cleaner: the model invokes add_to_cart, remove_from_cart, apply_promo_code as tools, rather than trying to parse intent via regex. This halves recognition errors and reduces cart abandonment by 25%.

UI on the mobile client

Beyond text chat, the bot often uses structured elements:

Quick reply buttons — after 'Payment method?' we show options as tappable chips, so the user doesn't have to type.

Product cards — when confirming the order contents, we display mini-cards with image, name, price. On Android this is a RecyclerView with horizontal scroll inside a bubble message; on iOS — UICollectionView with horizontalScrollDirection.

Order summary — a final screen before confirmation as a separate component, not plain text. The user sees the full list, total, address, and a 'Place order' button. This UI boosts conversion by 15%.

Status notifications

After the order is placed, the bot continues working via push notifications: 'Order received', 'Courier on the way', 'Delivered'. On iOS — APNs via Firebase Cloud Messaging, on Android — FCM directly.

Notifications include a deep_link with parameters that open the chat history for that specific order, not the main app screen.

Process

1. Scenario design: regular order, modification, cancellation, reorder, promo code.
2. Server state machine development + integration with order API.
3. Mobile UI: bubble layout, quick replies, product cards, summary.
4. Testing edge cases: mid-process interruption, conflicting commands, expired sessions.

Technical details of the state machine: on the server it's implemented in Python using the transitions library. States and transitions are defined in a YAML config, allowing scenario changes without rebuilding. Each state has a TTL timeout — if the user is idle for 30 minutes, the session ends.

What is included in the deliverable

• Server and mobile client source code
• API and scenario documentation
• Push notification and deep linking setup
• Real-device testing (iOS/Android)
• Deployment guide for App Store and Google Play
• 6-month warranty on bugs

On iOS we use StoreKit 2 for in-app purchases and TestFlight for beta testing. The state-machine order bot is suitable for complex scenarios.

Timeline estimates

A bot with a linear scenario (selection → address → payment → confirmation) + mobile client: 1–1.5 weeks. With non-linear scenarios, loyalty system integration, order history, and push notifications: 3–4 weeks.

Our team has been in mobile development for over 10 years. Order an audit — we'll propose the optimal solution. Request a consultation — we'll explain how a bot fits your business.

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.