Architecture

Architecture

Helpdesk.AI utilizes a clean, decoupled architecture built for production SaaS environments. It is structured into multiple layers to ensure separation of concerns, scalability, and security.

1. System Architecture

graph TD
    A["User (Frontend)"] -->|"Submits Issue"| B("FastAPI Backend")
    B -->|"Text Processing"| C{"AI Inference Engine"}
    C -->|"DistilBERT v3 (Categorization)"| D["Database Routing"]
    C -->|"NER Engine (Entity Extraction)"| D
    C -->|"GitHub Models / GPT-4o (Reasoning)"| E["Auto-Resolution Suggestions"]
    D --> G[("Supabase DB")]
    E --> G
    G -->|"Real-time Sync"| A
    G -->|"Dashboard Data"| H["Admin/Agent Portal"]

Loading

2. The AI Neural Pipeline

Under the hood, Helpdesk.AI leverages a custom-orchestrated suite of models, now augmented with GitHub Models integration.

High-Precision Classification

Driven by DistilBERT v3, our classifier understands technical context and user sentiment to assign accurate Impact Scores (Low -> Critical).

NER Metadata Harvesting

Extracts crucial infrastructure identifiers:

• Assets: IP Addresses, Hostnames.
• Environment: OS, Browser types.

GitHub Models Integration

For complex reasoning, resolving, and summarization, we integrate directly with GitHub Models via Azure AI Inference. This allows us to instantly swap to the best LLMs (like gpt-4o) for tasks such as:

• Generating user-friendly resolution steps.
• Summarizing massive email threads into concise ticket descriptions.
• Generating knowledge base articles dynamically.

See scripts/github_models_bridge.py and .github/workflows/models-evaluation.yml for implementation details.