🚀 Live Demo (Please allow ~50 seconds for the initial wake-up)

🚀 PulseDesk: AI-Powered Multilingual Support Triage

PulseDesk is a modern backend service designed to solve a common enterprise problem: efficiently triaging customer feedback in a multilingual environment. Using state-of-the-art Large Language Models (LLMs), PulseDesk automatically translates, categorizes, and prioritizes incoming comments in real-time.

🌟 Key Features

• Multilingual Intelligent Triage: Native support for English, Russian, and Lithuanian.
• Automated Translation: Non-English comments are automatically translated into English for internal ticket management.
• High-Availability AI Strategy: Implements a "Model Fallback" chain to ensure 100% uptime even if a specific AI provider is down.
• Glass-Morphism UI: A sleek, modern dashboard for visualizing live comment-to-ticket conversion.
• API Documentation: Fully documented endpoints via Swagger/OpenAPI.

🛠️ Tech Stack

• Backend: Java 17, Spring Boot 4.0.2, Spring Data JPA.
• Database: H2 (In-Memory) for rapid demonstration.
• AI Integration: Hugging Face Inference API (Router).
• Frontend: HTML5, Tailwind CSS, JavaScript (Fetch API).
• Documentation: SpringDoc OpenAPI (Swagger).

🧠 Advanced Technical Implementation

AI Resilience Logic

PulseDesk does not rely on a single AI model. To handle the volatility of 2026 AI providers, I implemented a fallback mechanism in HuggingFaceService.java. If the primary model fails or returns invalid formatting, the system automatically tries the next candidate:

1. Qwen 2.5 7B (Together AI) - Primary for speed & multilingual accuracy.
2. Mistral 7B v0.2 (Featherless AI) - Secondary for stability.
3. Gemma 2 9B - Tertiary fallback.

Robust JSON Parsing

Large Language Models occasionally return "chatty" responses or missing delimiters. PulseDesk utilizes a custom Regex-based sanitization layer and a lenient Jackson 3 configuration to "peel" JSON content out of Markdown code blocks and repair missing root braces before persistence.

🚀 Getting Started

Prerequisites

• Java 17 JDK or higher
• Maven 3.6+
• A Hugging Face API Token (with access to the Inference API)

Installation

1. Clone the repository:

git clone https://github.com/m21tomas/pulsedesk.git

Navigate to the project directory

cd pulsedesk

2. Set your API Token: Set your Hugging Face access token as an environment variable:

export HF_API_TOKEN=your_token_here

3. Run the application: Windows:

mvnw spring-boot:run

Windows (Git Bash) / Linux / macOS

./mvnw spring-boot:run

or using installed Maven:

mvn spring-boot:run

4. Access the Dashboard: Open http://localhost:8080 in your browser.

📈 Testing the System

The application is pre-loaded via DataInitializer.java with 6 diverse test cases:

• English: Bug reports and general praise.
• Lithuanian: Feature requests and support team feedback.
• Russian: Complex billing issues and gibberish detection.

📝 API Endpoints

You can explore the full API schema via Swagger at: http://localhost:8080/swagger-ui/index.html