Architecture

Note

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 strict security isolation.

System Architecture The system is split horizontally: Frontend: React for user interactions. Backend: High-speed FastAPI routing. Database: Supabase with strictly controlled Row-Level Security (RLS). Intelligence: The custom AI Inference Engine handles Text Processing and Reasoning as side-effects before database commits.

```mermaid graph TD A["User (React UI)"] -->|"Submits API Req"| B("FastAPI Backend") B -->|"Text Payload"| C{"AI Inference Engine"} C -->|"DistilBERT v3"| D["Categorization Array"] C -->|"NER Engine"| D C -->|"GitHub Models / GPT-4o"| E["Auto-Resolution Suggestions"] D --> G[("Supabase Storage")] E --> G G -->|"Real-time WebSocket Sync"| A G -->|"Dashboard Aggregation"| H["Admin Portal"] ```

Important

The AI Neural Pipeline

Helpdesk.AI leverages a custom-orchestrated suite of models, natively augmented with GitHub Models integration to separate logic layers.

High-Precision Classification Driven by DistilBERT v3, our classifier understands deep technical context and user sentiment to assign accurate Priority Impact Scores without hallucinating.

Metadata Harvesting A custom NER pipeline extracts crucial infrastructure identifiers automatically from plain text (e.g., Hostnames, IP Addresses, Browser types) eliminating manual forms.

GitHub Models Integration For complex logic like generating resolution steps or summarizing massive email threads, we pivot to GitHub Models via Azure AI (specifically `gpt-4o`) to ensure conversational accuracy.