Mobile App Booking Bot Development for Restaurants and Hotels

Replacing multi-step forms with natural dialogue is a key task when booking tables or rooms. A user writes: "A table for two, tomorrow at 7 PM, by the window, non-smoking area" — and the bot instantly extracts all parameters, checks availability via PMS, and offers a slot. No three separate pickers. This approach reduces booking time by 30% and cuts abandoned sessions by 15%. We have deployed over 10 such solutions for restaurants and hotels, and guarantee stable performance under load. Development cost depends on integration complexity, but the savings in operator time pay back the investment in 3–5 months.

How the bot understands the request and extracts parameters

For table booking, typical slots are: date, time, party size, zone preference (terrace, main hall, bar), occasion, guest name. The bot must recognize them in a single phrase. We use two approaches.

Classic NLU: Dialogflow CX with system entities @sys.date-time and @sys.number covers basic cases. For Rasa, we use duckling as an entity extractor plus custom entities for zone types. Extraction accuracy in test scenarios: 92%.

LLM with structured output: When the request is more complex or high accuracy is needed, we apply a model with JSON Schema. For example, based on OpenAI Structured Outputs:

from openai import OpenAI
from pydantic import BaseModel

class BookingSlots(BaseModel):
    date: str | None = None      # ISO 8601
    time: str | None = None      # HH:MM
    party_size: int | None = None
    zone_preference: str | None = None
    guest_name: str | None = None
    occasion: str | None = None

response = await client.beta.chat.completions.parse(
    model="gpt-4o-mini",
    messages=[
        {"role": "system", "content": "Extract booking parameters from the user's text."},
        {"role": "user", "content": user_message}
    ],
    response_format=BookingSlots
)
slots = response.choices[0].message.parsed

The model returns only the fields present in the message. The bot asks for missing ones one by one — that's natural dialogue, not a form. With LLM, accuracy reaches 97% — 5 percentage points better than classic approaches. However, LLM requires more compute resources, so for simple scenarios we use Dialogflow CX.

Why a bot is more effective than a form

A form with pickers forces the user to select each parameter sequentially, which is annoying and increases time. A bot with NLU processes the request in one step — reducing cognitive load. In one project for a coffee chain, we replaced a 5-step form with a dialogue, and booking conversion increased by 25%. Average booking time dropped from 45 to 12 seconds.

Real-time availability check

Before showing a slot, we must ensure it is free. We integrate with the venue's booking system:

• Restaurants: iiko, r_keeper, Tillypad — each has a booking API.
• Hotels: Opera PMS, Fidelio, Apaleo (via Channel Manager).
• Custom systems: REST API with an available slots endpoint.

It's important to return not just "free/occupied" but a list of alternatives. If the requested time is taken, the bot offers 3–5 nearest available slots. This increases confirmation conversion by 20%.

How to avoid double booking?

Between "slot shown" and "user confirmed", 2–3 minutes pass. During that time, another guest might take the slot. Solution: optimistic locking with a short TTL. When showing a slot, send PUT /reservations/hold with a 3-minute TTL. On confirmation, send POST /reservations/confirm. If the user doesn't confirm, the hold is automatically released.

We show a countdown timer: "Table reserved for 3:00" — this reduces anxiety and speeds up decision-making by 20% compared to locking without a timer. We implement it on the mobile client:

// Android: countdown timer
class BookingViewModel : ViewModel() {
    private var holdExpiresAt: Long = 0

    fun startHoldCountdown(ttlSeconds: Int) {
        holdExpiresAt = System.currentTimeMillis() + ttlSeconds * 1000L
        viewModelScope.launch {
            while (System.currentTimeMillis() < holdExpiresAt) {
                val remaining = (holdExpiresAt - System.currentTimeMillis()) / 1000
                _holdCountdown.emit(remaining)
                delay(1000)
            }
            _holdExpired.emit(Unit)
        }
    }
}

UI components for mobile app

Modifying and cancelling bookings is also done through the bot: the user types "cancel booking" or "reschedule for tomorrow". The bot recognizes the command, finds the active booking by account, calls the API, and confirms the change.

What's included in the work

1. Analysis of the venue's booking system, API documentation.
2. Dialogue design: mandatory and optional slots, alternatives when occupied.
3. Server-side development: PMS/booking API integration, hold logic.
4. Mobile UI: dialogue with inline components, confirmation card, calendar.
5. Push notification setup for confirmation and reminders.
6. User-side testing and stress testing under concurrent requests.
7. Integration documentation and staff training.
8. Code warranty for 3 months after deployment.

Timeline estimates

Contact us — we'll assess your project and provide exact timelines. Our experience: over 10 successfully deployed booking bots for restaurants and hotels. Get a consultation — we'll tell you how to reduce booking time and increase conversion.

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.