NER in Mobile Applications: Extracting Entities from Text

Extracting Named Entities (NER) in a Mobile Application

Our NER mobile app solutions integrate named entity recognition for entity extraction on iOS and Android. We utilize spaCy NER for lightweight tasks and Hugging Face NER for high accuracy, with custom NER model training for domain-specific entities. The hybrid NER regex approach combines TFLite NER on-device with server-side APIs for smart form autofill and NER API integration across platforms. This enables efficient text processing in apps and seamless NER model training.

In a delivery app, a user types "deliver on Thursday at 14:00 to Lenina 5 apt 12" into a single field. Without NER, you'd need dozens of regular expressions — and every non-standard format breaks the logic. We solve this with a custom NER model: it transforms free-form text into structured fields {date, time, street, building, apartment}. Input time savings — up to 40%, and error rate drops by 60%. Basic integration costs $2,000–$5,000, and full integration with custom models is $10,000–$20,000. Basic integration takes 3 to 5 days, full deployment with smart forms — 2 to 3 weeks. Cost is calculated individually for your case.

How NER Converts Text to Structured Data

NER (Named Entity Recognition) is an NLP task that extracts entities from text: names, locations, dates, organizations. In mobile apps, NER is used for:

• Smart forms and autofill. The user writes a message to the courier — the app parses the address and delivery time without separate form fields.
• Search with filters. "iPhone 15 Pro 256GB black" -> {brand: Apple, model: iPhone 15 Pro, storage: 256GB, color: black}. A structured query is more accurate than full-text search.
• Chatbots and voice assistants. Extraction of parameters from free speech or text to fill dialog slots.
• Receipt and document processing. OCR text from a receipt -> {store, amount, date, items}.

Why Choose a Hybrid Approach?

Pure transformer-based NER provides high accuracy but is slow and requires a server. Pure regex is fast but breaks on free text. On-device NER is up to 10x faster than server API, with zero latency and no network costs. A hybrid regex+NER approach combines the speed of rule-based extraction for standard entities (phones, emails, SKUs) with ML flexibility for complex contexts. In most projects, this is the optimal balance of performance and quality.

Technical Approaches for Mobile Applications

Comparison of on-device vs server-side NER:

spaCy + Custom NER Model

spaCy is a standard for production NER. The base Russian model ru_core_news_lg recognizes persons, organizations, locations, dates. For domain-specific entities (clothing sizes, SKUs, medical terms), fine-tuning is required. We use BIO tagging scheme and conditional random fields (CRF) for sequence labeling.

import spacy
from spacy.training import Example

# Load base Russian model
nlp = spacy.load("ru_core_news_lg")

# Add custom entity types
ner = nlp.get_pipe("ner")
ner.add_label("PRODUCT_SIZE")
ner.add_label("PRODUCT_COLOR")
ner.add_label("ARTICLE")

# Training example
TRAIN_DATA = [
    ("I want to find sneakers size 42 in blue color SKU 98765",
     {"entities": [(33, 35, "PRODUCT_SIZE"), (43, 49, "PRODUCT_COLOR"), (59, 64, "ARTICLE")]}),
]

optimizer = nlp.resume_training()
for text, annotations in TRAIN_DATA:
    doc = nlp.make_doc(text)
    example = Example.from_dict(doc, annotations)
    nlp.update([example], sgd=optimizer)

Official spaCy NER training documentation: https://spacy.io/usage/training#ner

Transformer-based NER via Hugging Face

For high accuracy on complex domains, we use a fine-tuned DeepPavlov/rubert-base-cased with a token classification head and attention mechanisms. This approach is slower than spaCy but handles contextual dependencies significantly better.

from transformers import pipeline

ner_pipeline = pipeline(
    "token-classification",
    model="DeepPavlov/rubert-base-cased-ner",
    aggregation_strategy="simple"
)

def extract_entities(text: str) -> list[Entity]:
    raw_entities = ner_pipeline(text)
    return [
        Entity(
            text=e["word"],
            label=e["entity_group"],
            confidence=e["score"],
            start=e["start"],
            end=e["end"]
        )
        for e in raw_entities
        if e["score"] > 0.7
    ]

How to Implement a Hybrid Extractor?

import re
from typing import NamedTuple

class EntityExtractor:
    PHONE_PATTERN = re.compile(r'(?:\+7|8)[\s\-]?\(?\d{3}\)?[\s\-]?\d{3}[\s\-]?\d{2}[\s\-]?\d{2}')
    EMAIL_PATTERN = re.compile(r'\b[A-Za-z0-9._%+-]+@[A-Za-z0-9.-]+\.[A-Z|a-z]{2,}\b')
    DATE_PATTERN = re.compile(r'\b(\d{1,2})[./](\d{1,2})(?:[./](\d{2,4}))?\b')

    def extract_all(self, text: str) -> dict:
        phones = self.PHONE_PATTERN.findall(text)
        emails = self.EMAIL_PATTERN.findall(text)

        ner_entities = extract_entities(text)
        locations = [e.text for e in ner_entities if e.label in ("LOC", "GPE")]
        persons = [e.text for e in ner_entities if e.label == "PER"]

        return {
            "phones": phones,
            "emails": emails,
            "locations": locations,
            "persons": persons
        }

Mobile Integration

iOS: NER for Smart Form Filling

// iOS: NER via server API with form autofill
class AddressFormViewModel: ObservableObject {
    @Published var street = ""
    @Published var building = ""
    @Published var apartment = ""
    @Published var deliveryTime = ""

    func parseFromText(_ userText: String) {
        Task {
            let entities = try await nerApi.extract(text: userText)

            await MainActor.run {
                if let address = entities.first(where: { $0.label == "ADDRESS" }) {
                    parseAddressComponents(address.text)
                }
                if let time = entities.first(where: { $0.label == "TIME" }) {
                    deliveryTime = time.text
                }
            }
        }
    }
}

On-device NER via CoreNLP or TFLite

For simple domain entities (SKUs, sizes, colors), you can deploy a compact TFLite NER model (< 20 MB) directly on the device. This removes latency and works offline.

Apple's NaturalLanguage.framework with NLTagger supports basic entity types out of the box for Latin texts:

let tagger = NLTagger(tagSchemes: [.nameType])
tagger.string = userInput

tagger.enumerateTags(in: userInput.startIndex..<userInput.endIndex,
                     unit: .word,
                     scheme: .nameType,
                     options: [.omitWhitespace]) { tag, range in
    if let tag = tag {
        print("Entity: \(userInput[range]), type: \(tag.rawValue)")
    }
    return true
}

For Russian, NLTagger performs noticeably worse — we use it only as a prefilter or for apps with Latin text.

What's Included in the Work

We offer a full-cycle NER integration into your mobile app:

• Analysis of the subject area and definition of target entities.
• Collection and annotation of training data (from 500 examples per entity).
• Architecture selection: spaCy, transformer, or hybrid.
• Model development and training.
• Integration into the mobile client (iOS/Android) via API or on-device.
• Creation of smart UI components: form autofill, entity highlighting, search filters.
• Testing on real data and latency optimization.
• Deliverables: documentation, model artifacts (spaCy pipelines or TFLite files), API endpoints for NER, mobile SDK (Swift/Kotlin wrappers), training sessions for your team, and 3 months of support.

Timeframe Estimates

We follow a clear integration process:

1. Analysis of subject area and entity definition — 1 day.
2. Data collection and annotation — 1 week.
3. Architecture selection — 1 day.
4. Model development and training — 1 week.
5. Mobile integration — 1 week.
6. UI components creation — 3 days.
7. Testing and optimization — 2 days.
8. Documentation and training — 2 days.

• Basic NER with a ready Russian model + API — 3–5 days.
• Fine-tuning on custom entities — 1–2 weeks.
• Full integration into mobile UI (smart forms, search, chatbot) — 2–3 weeks.
• On-device model (TFLite/CoreML) — additional 1–2 weeks.

We have over 6 years of experience in mobile development and have completed more than 30 NLP projects. We guarantee high accuracy with certified models and provide a warranty on all deliverables. If you need NER integration into your mobile app — contact us for a project assessment. We will calculate the cost and timeline for your case and suggest the optimal solution. Get a free consultation.

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.