Integrating Dialogflow NLP into a Mobile Chatbot

Integration of NLP Engine (Dialogflow) into a Mobile Chatbot

We often see projects where Dialogflow integration starts without a clear understanding of the differences between ES and CX. Clients choose ES for its simplicity, but six months later face the need to rewrite all intent logic—the State Machine architecture in CX is incompatible with the linear Flow from ES. Our task is to help you choose the right version and configure the integration so that the bot works stably from the first release. With over 5 years in mobile development, we guarantee quality at every stage.

Why the Choice of Dialogflow Version Is Critical

Dialogflow CX and Dialogflow ES are fundamentally different products. ES uses a linear flow: intents are processed sequentially, contexts live for a limited number of turns. CX is built on a State Machine: each step is a page with explicit transitions. Migrating from ES to CX requires a complete redesign of the agent. CX is better suited for complex multi-step scenarios (order placement, tech support), while ES is for simple Q&A. The choice depends on your scenarios; we offer an audit before development begins.

How to Set Up Authentication Without Security Risks

The official Google Cloud documentation suggests using a service account JSON directly in the app. This is unacceptable—the private key ends up in the APK/IPA. The correct approach: the mobile app sends text to your backend, and the backend communicates with Dialogflow via the google-cloud-dialogflow library using the service account. We implement a proxy server in Node.js or Python that processes requests and returns responses. Example in Node.js:

const { SessionsClient } = require('@google-cloud/dialogflow-cx');
const client = new SessionsClient();

async function detectIntent(projectId, location, agentId, sessionId, text, languageCode) {
  const sessionPath = client.projectLocationAgentSessionPath(
    projectId, location, agentId, sessionId
  );
  const request = {
    session: sessionPath,
    queryInput: {
      text: { text },
      languageCode,
    },
  };
  const [response] = await client.detectIntent(request);
  return response.queryResult;
}

How to Solve Multilingual and Context Issues

Dialogflow supports multiple languages on one agent, but training phrases for each language must be added separately. If you pass languageCode: ru to an agent without Russian phrases, Dialogflow returns a fallback intent with low confidence—the user won't understand the answer. We configure language sets and test them via the simulator.

Context management is a common source of errors. In ES, contexts live for N turns after being set. If you don't reset the context after a scenario ends, the user might get an unexpected response. In CX, this issue is solved by explicit transitions between pages. Our engineers design flows so that each scenario ends with a context reset for all active dialogues.

Mobile-Side Implementation

On Android, we use OkHttp or Retrofit to call the proxy server. Session ID is generated once at session start and kept in memory—not in SharedPreferences, sessions should not survive app restarts.

class DialogflowRepository(private val api: ChatApiService) {
    private val sessionId = UUID.randomUUID().toString()

    suspend fun sendMessage(text: String, locale: String): ChatResponse {
        return api.detectIntent(
            DetectIntentRequest(
                sessionId = sessionId,
                text = text,
                languageCode = locale
            )
        )
    }
}

On iOS—similarly via URLSession or Alamofire. No direct calls to Google API from the client. For rendering Rich responses (cards, buttons, carousels), we parse fulfillmentMessages and render the corresponding view components: TextBubbleCell, ButtonsRowCell, CardCell.

Agent Design: Practical Recommendations

For a production bot, you need:

• Group intents by domain (orders, support, FAQ) and cluster them
• Configure a fallback intent with multiple response variants (at least 5)
• Add Small Talk as a separate flow—otherwise the bot rudely ignores informal remarks
• Webhook fulfillment for dynamic responses (order status, stock levels)
• Training phrases: at least 10–15 variants per intent, otherwise confidence will be low on natural speech. We use automatic synonym generation to increase coverage.

Comparison of Dialogflow ES and CX

CX outperforms ES for multi-step scenarios by at least 2x in development speed and stability. Budget savings on rewriting—up to 50% with the correct choice from the start.

What Our Work Includes

• Audit of current scenarios and requirements gathering
• Dialogflow version selection (ES or CX) with justification
• Agent design: intents, entities, contexts, flows/pages
• Proxy server development (Node.js/Python) with authentication
• SDK integration into the mobile app (Android Kotlin, iOS Swift)
• Webhook fulfillment setup with your CRM/database
• Testing via Dialogflow Simulator and on real devices
• Maintenance documentation (access, scripts, configs)
• Team training on agent management

How We Work: Stages

1. Analytics — break down scenarios, collect dialogue examples, identify edge cases.
2. Design — draw dialogue maps and get client approval.
3. Implementation — create the agent, write the proxy, integrate into the client.
4. Training and Testing — at least 50 test dialogues per language.
5. Deployment — publish to App Store / Google Play, set up monitoring.

Timeline Estimates

• Integration with an existing agent: from 3 to 4 days (including proxy and client setup).
• Development of an agent from scratch for 5–10 scenarios plus mobile client: from 1.5 to 2 weeks.
• Post-launch support: on request, with a one-month warranty for debugging discovered issues.

Contact us for a project evaluation—we'll select the optimal turnkey solution.

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.