NeuroDesk: AI Helpdesk Automation
A full-stack helpdesk experiment that connects text classification, API inference, and a support dashboard.
Problem
How can a text-classification model be connected to a helpdesk workflow so that predictions remain visible to support users?
NeuroDesk is a full-stack ticket-classification system. The backend exposes inference through an API, while the frontend presents ticket and analytics views. The available page documents the architecture; it does not establish a measured reduction in triage work or a validated accuracy level.
System Architecture
The system operates as a decoupled full-stack application:
1. The AI Engine (Backend)
Built with Django REST Framework and PyTorch, the backend serves as the brain of the operation.
- Hybrid Models: Utilizes both LSTM (Long Short-Term Memory) for deep semantic understanding and FastText for rapid baseline classification.
- API Wrapper: Exposes endpoints for real-time inference, allowing the frontend to request predictions on-the-fly.
2. The Interactive Dashboard (Frontend)
A modern SPA built with React and Material UI.
- Real-time Analytics: Visualizes ticket trends and classification confidence using ApexCharts.
- Role-Based Access: Specialized views for Admins, Agents, and Users.
- State Management: Robust data handling with Redux Toolkit.
Key Features
- Automated Triage: Tickets are tagged and routed instantly upon creation.
- Confidence Scoring: The interface can present model confidence so that an agent can review ambiguous cases. A confidence value is not the same as calibrated certainty.
- Performance Metrics: The dashboard includes views for operational and classification metrics; the page does not claim a benchmark result.
Product value and limits
The product value is the handoff between prediction and human work. A ticket can be classified by the backend, surfaced in the dashboard, and reviewed by a support role when the prediction is uncertain. The separation also lets the model service evolve without rebuilding the dashboard.
The page documents the integration points and interface responsibilities. It does not establish production accuracy, calibrated confidence, or improved resolution time.
Technical Stack
| Component | Technology |
|---|---|
| Frontend | React, Redux, Material UI, ApexCharts |
| Backend | Python, Django REST Framework |
| ML Core | PyTorch, FastText, Pandas, NLTK |
| Database | SQLite / PostgreSQL |