Mobile Business Card Scanner App Development with OCR

The task looks simple: take a photo of a business card and get a contact in your address book. In practice, between the shot and a correctly populated CNContact, there's a chain where everything that can break breaks: poor lighting, non-standard fonts, bilingual cards, vertical text on Japanese cards. We develop turnkey business card scanners that handle these issues and deliver over 95% recognition accuracy.

What business card scanning gives to your business

Business card scanning solves three tasks: instant contact entry without manual typing, automatic CRM enrichment, and elimination of human error. On average, one contact is entered in 3 seconds instead of 30 — saving 27 seconds per card. For a manager processing 50 cards per week, this frees over 20 hours per year. At an average hourly rate of $30, that's $600 savings per month. Development cost starts at $5,000 with payback in 2-3 months.

Problems we solve

1. Image quality. Most users shoot cards carelessly: at an angle, while moving, in poor light. Our system automatically detects the card via VNDetectRectanglesRequest (iOS) or ObjectDetector (ML Kit), corrects perspective (CIPerspectiveCorrection), enhances contrast, and captures only when the card occupies >60% of the frame and is stable for 0.5 seconds. We don't use manual capture — it yields 30% more rejects.

2. Text recognition in difficult conditions. On iOS we use Apple's VNRecognizeTextRequest (offline, 18 languages). On Android — ML Kit Text Recognition v2 (offline, Latin + Cyrillic + CJK). For maximum accuracy, we connect Google Cloud Vision or AWS Textract — they return structured output with bounding boxes. Comparison of approaches:

Vision is faster and simpler, but on complex fonts Cloud API wins by 2-3%. The choice depends on offline requirements.

3. Parsing unstructured text. OCR outputs an array of lines, not contact fields. We use regular expressions for phones and emails, and NER (CoreML for iOS, nlp in ML Kit for Android) for names and positions. Bilingual cards (Russian/English) are processed separately: we determine the language of each line via NLLanguageRecognizer and apply different parsing rules.

To boost accuracy, we combine deterministic rules (phones, email) with trained NER models. On iOS, we use a CoreML model trained on a corpus of 10,000 business cards. On Android, ML Kit Entity Extraction supports 6 categories. This allows correct extraction of names even in non-standard formats (e.g., "John M. Smith Jr.").

How we achieve 95%+ accuracy

Accuracy is ensured by a combination of stages:

For critical cases, we recommend a hybrid approach: on-device for quick preview, cloud for final verification. According to Apple Vision documentation, on-device recognition achieves up to 98% accuracy under good lighting. Our scanner is 3x faster than manual entry — 5 seconds vs 30 seconds per card.

From our practice: a case study

We implemented a scanner for a trading company with 150+ managers. Stack: iOS — SwiftUI + Vision + CoreData, Android — Jetpack Compose + ML Kit + Room. A particular complexity was integration with AmoCRM via REST API. After scanning, the manager confirms fields in the UI, and data lands in a deal within 2 seconds. Manual entry is eliminated. Result: processing speed per card — 5 seconds, saving 25 hours per month per manager. Development investment paid off in 2 months.

Our process

1. Audit: which languages, offline need, integration with CRM or just device contacts.
2. Design: module architecture (capture → correction → OCR → parsing → editing → export).
3. Implementation: we code following best practices (perspective correction, auto-capture, NER).
4. Testing: on a set of 100+ real business cards of varying quality and formats.
5. Deployment: publish to App Store / Google Play, configure push notifications (APNs/FCM), deep linking.

What's included in the work

• Source code of the app (iOS/Android/Flutter) with documentation.
• Repository access (Git) and CI/CD (GitHub Actions / GitLab).
• Two testing iterations with our business card set.
• Training of the client's team on admin panel and export functions.
• 3-month warranty on recognition-related bugs.

Timeline estimates

Scanner with Vision/ML Kit, basic parsing, and saving to contacts — 2–3 weeks. With cloud OCR, NER, multi-language support, and CRM integration — 5–7 weeks. We refine timelines during the audit — get a free consultation.

Typical mistakes in self-development

• No auto-capture: user takes photo with shaky hands → blurry image.
• Using only regex for parsing: misses names in non-standard format.
• Ignoring perspective correction: tilted text is recognized 20% worse.

Our implementation leverages lemmatization and tokenization techniques to boost parsing accuracy. Result: 95%+ accuracy in any conditions.

Contact us to discuss your project. We have 5+ years in mobile development and over 50 projects in OCR and scanning. We guarantee the result — we'll specify recognition accuracy in the TOR.

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.