AI System

Previous: API Overview
Next: SLA System

The AI system supports intake and triage by classifying issue text and media, then storing the resulting prediction data with each request.

1) Runtime ownership

• Runtime inference and AI-integrated endpoints are in smart-estate-backend.
• Training and artifact generation are in smart-estate-ai.

2) Runtime flow

Message and/or media input
  -> text analysis (chatbot.py)
  -> vision analysis (OpenAI Vision first, with local model support when available)
  -> fusion (fusion_engine.py)
  -> ticket updates + AI decision log persistence
  -> assignment + notifications

3) Implemented behavior

• POST /api/chat handles AI-assisted intake and can create tickets.
• POST /api/upload-image supports media AI path.
• PATCH /api/ai/logs/{log_id}/feedback records corrected categories.
• GET /api/ai/status reports runtime AI availability.

4) Data captured

• Ticket-level AI result data (ai_result).
• AI decision records (ai_logs) with predicted category, confidence, and feedback fields.
• Metrics endpoint aggregates AI confidence and feedback coverage.

5) Runtime considerations

• Runtime quality depends on configured provider keys and available local artifacts.
• Tenant mobile AI interactions currently focus on guided intake workflows, while broader ticketing and chat capabilities are handled through existing backend-driven flows.

Caption: AI confidence is surfaced alongside ticket and SLA metrics in current dashboard UI.

Previous: API Overview
Next: SLA System