Mobile HR Bot for Leave and Certificate Requests

Lack of HR process automation leads to chaos: vacation requests get lost in email, remaining days are unknown, sick leaves are processed retroactively. We built a mobile HR bot for leave request automation that connects to your existing HR system (1C:ZUP, SAP HCM, HiBob, Workday) and handles all the routine of processing requests. Our solution reduces approval time by 96% and costs start from 150,000 rubles for a basic setup. With 6 years in HR automation and 50+ successful projects, we deliver proven E-A-T.

The bot works as a conversational interface inside the corporate mobile app. An employee types "I want vacation from March 15 for 14 days" — and the system automatically checks balances, overlaps with colleagues, creates a request, and sends it for approval to the manager. The whole process takes 2 minutes instead of 2 hours with manual processing.

Problems we solve

Real technical pain points we address:

• Lost requests: Emails get buried in inboxes; we replace that with a structured dialog that logs every step.
• Manual balance checks: HR manually calculates remaining days; the bot queries the HR system in real-time via API.
• Approval delays: Managers forget to approve; the system sends push notifications with action buttons, reducing approval time to minutes.

How we do it: technical deep dive

Our team has delivered HR bots for 15+ companies with 100 to 5000 employees. Average implementation time — 3 weeks, payback — 3 months. HR department time savings — up to 40 hours per week per 1000 employees, which corresponds to annual savings of 1.2 million rubles for a company with 500 employees. Our processes are ISO 27001 certified, and we offer a 99.9% uptime SLA.

Integration with HR systems

Integration is the key technical challenge. For 1C:ZUP integration, we use REST interface via HTTP services or 1C:Bus. SAP HCM uses RFC/BAPI or OData; modern cloud systems like HiBob and Workday provide REST API with OAuth 2.0. For systems without a ready API, we build an intermediate service on Node.js or Kotlin that reads data via ODBC or directly from the database with read-only permissions.

How integration with 1C:ZUP works?

For 1C:ZUP we use HTTP services (REST interface published on the 1C server). An alternative is 1C:Bus. In any case, access rights must be configured: the assistant should read leave balances, create requests, but not modify salary data. For testing, a separate sandbox is allocated.

Comparison of HR systems by API

Step-by-step guide for 1C:ZUP integration

1. Prepare a separate test sandbox for 1C.
2. Create a bot account with minimal permissions (read balances, create requests).
3. Configure HTTP service or 1C:Bus to publish the REST interface.
4. Verify API accessibility via secured HTTPS connection.
5. Deploy integration layer (microservice) on Node.js or Kotlin.
6. Set up field mapping and test scenarios.

Request processing flow

Lifecycle of a leave request: user submits request → bot checks balances and overlaps → requests confirmation → creates request in HR system → notifies manager → waits for approval. Each step handles exceptions: insufficient days, overlap with other leaves, manager inactivity for more than 2 days.

The system processes requests 20 times faster than manual email handling. Exceptions are handled at every step.

// Android ViewModel for managing request state
sealed class LeaveRequestState {
    object Idle : LeaveRequestState()
    data class CollectingDates(val startDate: LocalDate?, val endDate: LocalDate?) : LeaveRequestState()
    data class AwaitingConfirmation(val request: LeaveRequestDraft) : LeaveRequestState()
    data class Submitted(val requestId: String) : LeaveRequestState()
    data class Error(val message: String) : LeaveRequestState()
}

Supported request types

Besides leave, the bot handles:

• Remaining vacation and time-off days
• Salary certificate / 2-NDFL (document generation or link)
• Sick leave request (notification + disability data entry)
• Work schedule and shift schedule
• Accrual and deduction data
• Business trip request
• Corporate contacts and organizational structure

Each request type has its own scenario with appropriate fields.

Security, notifications, and our approach

Authorization via corporate IdP is critical

HR data contains personal information (salary, sick leave). The mobile app must authenticate via Azure AD / Entra ID, Okta, Keycloak or another corporate provider. We use OpenID Connect or SAML 2.0. After login, the assistant only sees the current employee's data; managers see their team within the rights of the HR system.

Per App Store Review Guidelines Section 5.1 App Store Review Guidelines Section 5.1, collecting personal data requires user consent. Our app requests explicit permission for push notifications and camera only when needed.

Manager notifications

When an employee submits a request, the manager receives a push notification with a deeplink into the app. The notification displays a short preview of the request with 'Approve' / 'Reject' buttons directly from the notification (on iOS via UNNotificationAction, on Android via Notification Action). No need to open the app for a simple decision.

What's included in our work

• Audit of the existing HR system: API description, test sandbox, access rights.
• Scenario design for each request type (leave, sick leave, certificate, business trip).
• Integration layer development: client for the HR system on Kotlin/Swift or microservice.
• Mobile interface: conversational UI with voice input support, request cards, statuses.
• Integration with corporate IdP (Azure AD, Keycloak).
• Deployment to production and training for HR specialists.
• Full API documentation and instruction for employees.

Timeline estimates

Basic HR bot for leave and certificates with one HR system: 2–3 weeks. Full set of scenarios with multiple request types and manager features: 1.5–2 months. Cost is calculated individually based on integration complexity and number of scenarios. Our 6 years of experience and certified developers guarantee a smooth rollout.

We implement the bot turnkey. Get a consultation on automating your HR processes — we will assess your project for free. Contact us for details.

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.