Rasa NLP Integration for Mobile Chatbots

Standard Rasa config on Russian yields about 50 % accuracy — on short phrases, this means every second request is misrecognized. Clients with strict confidentiality requirements (medicine, finance) cannot use cloud NLP services like Dialogflow or Google NL. For them, Rasa is the only option, but it needs proper tuning. We have been doing this for many years: over 20 projects, average accuracy improvement from 50% to 85% through custom pipeline and high-quality dataset.

Why Rasa is Better than Dialogflow for Sensitive Data

Rasa is self-hosted, Dialogflow is cloud-based. For companies where data must stay on-premise, the difference is fundamental. Rasa gives full control over the pipeline: you can replace any component, add your own tokenizer, use custom models. Dialogflow limits you to built-in mechanisms. At scale of 10 000+ dialogues per day, Rasa is 3–4 times cheaper due to no per-request fees.

How We Configure Rasa for Russian

The main issue is that the default WhitespaceTokenizer works poorly with Russian morphology. We replace it with SpacyNLP using the ru_core_news_md model and add char n-gram features via CountVectorsFeaturizer. Here is the production config we use:

language: ru

pipeline:
  - name: SpacyNLP
    model: ru_core_news_md
  - name: SpacyTokenizer
  - name: SpacyFeaturizer
  - name: RegexFeaturizer
  - name: LexicalSyntacticFeaturizer
  - name: CountVectorsFeaturizer
  - name: CountVectorsFeaturizer
    analyzer: char_wb
    min_ngram: 1
    max_ngram: 4
  - name: DIETClassifier
    epochs: 150
  - name: EntitySynonymMapper
  - name: ResponseSelector
    epochs: 100
  - name: FallbackClassifier
    threshold: 0.7
    ambiguity_threshold: 0.1

This pipeline yields 85 % accuracy on a set of 20–50 intents. DIETClassifier with 150 epochs trains in 5–10 minutes on a VPS with 4 GB RAM.

How Rasa Core Manages Dialogue

Rasa separates NLU (intent recognition) and Core (dialogue management). Core operates on rules (rules.yml) and stories (stories.yml). The most common mistake is trying to describe all scenarios via rules and ending up with a fragile system that breaks on non-standard utterance order.

Rule: rigid commands (cancel, help, restart) go in rules. Multi-step scenarios with variability go in stories. Rasa Core learns from stories and generalizes unseen scenarios — that is its main advantage over decision trees.

Custom Actions. Dynamic responses (order status, available slots) are implemented via action_server — a separate Python service that Rasa calls over HTTP. The mobile app does not interact directly with it:

class ActionCheckOrderStatus(Action):
    def name(self) -> str:
        return "action_check_order_status"

    async def run(self, dispatcher, tracker, domain) -> list:
        order_id = tracker.get_slot("order_id")
        status = await order_service.get_status(order_id)
        dispatcher.utter_message(text=f"Your order #{order_id}: {status}")
        return [SlotSet("order_status", status)]

How to Integrate Rasa with a Mobile App

Rasa Server provides a REST channel at /webhooks/rest/webhook. The mobile app sends a POST request:

{
  "sender": "user_device_id_or_session_uuid",
  "message": "message text"
}

The response is an array of messages, each can be text, image, buttons, or custom payload.

On Android, this is a standard Retrofit call. Important: sender must be a stable session identifier — Rasa stores slots between requests within the same sender. Generating a new ID each time will lose dialogue context.

For production, Rasa should not be exposed directly to the internet — we place nginx in front with rate limiting and token authentication.

Deployment

Rasa Server + Action Server are easily run via Docker Compose. The model is trained with rasa train and mounted into the container. On a small VPS, training 50 intents takes 5–10 minutes — perfectly acceptable for CI/CD.

Rasa Enterprise (commercial version) adds analytics and A/B testing of dialogues, but for most tasks the open-source version is sufficient.

What's Included in Our Work

• Domain audit and collection of example utterances for the NLU dataset.
• Pipeline configuration, base model training, accuracy evaluation via rasa test.
• Development of dialogue scenarios (rules + stories), custom actions.
• REST channel integration with the mobile client, infrastructure setup.
• Documentation and training for your team.
• 30-day warranty support.

Timeframes

Integration with an existing Rasa server — 3–4 days. Full cycle including model training from scratch, scenario writing, and deployment — 2–4 weeks depending on the number of intents.

Contact us for a project assessment. Get a consultation on configuring Rasa for your task.

Rasa Documentation | Wikipedia: Rasa

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.