AI Customer Operations — Auto Body Shop

The Challenge

A busy shop needed to handle high-volume inbound inquiries on WhatsApp — including photo-based damage assessments — without the owner answering every message. Leads were slipping through. Response times were inconsistent. There was no system, just a phone.

The Solution

Built a custom AI system integrated directly into the shop's WhatsApp channel via CRM. The system runs an 11-state conversation flow: greeting new clients, gathering vehicle and damage information, receiving and analyzing photos, routing complex cases to a human, and following up through payment. Multi-model routing selects the right AI for each turn — fast for simple exchanges, more capable for damage assessment and tool execution.

• FastAPI backend — 10,000+ lines of production Python handling message routing, AI orchestration, state management, and CRM sync
• 11-state conversation state machine — tracks each client through intake, assessment, quoting, editing, payment, and escalation
• Multi-model routing — lightweight model for simple turns, advanced model for complex analysis and tool calls
• 16-intent detection system — regex + LLM classification to route insurance disputes, complaints, and urgent cases instantly
• Vision analysis pipeline — customers send damage photos via WhatsApp, AI evaluates condition and generates preliminary assessments
• CRM integration — all conversations logged, leads tracked, quote status synced in real time
• WebSocket direct channel to quoting API — submissions and status checks without extra hops
• Telegram HITL bot — human-in-the-loop escalation for edge cases, with manual override and monitoring