ClaimsIQ: Car Insurance AI Agent
Car insurance claims agent using Claude Vision to classify vehicle damage from photos. Agentic pipeline from photo upload to structured coverage assessment.
Overview
An agentic prototype for automated car insurance claims processing. Claimants upload vehicle damage photos which Claude Vision classifies — detecting damage type, severity, and affected components — then an LLM-orchestrated pipeline applies coverage rules and produces a structured assessment with confidence scores. Built to explore where multimodal AI holds up in regulated claims workflows, and where confidence thresholds break down.
Challenge
Car insurance damage assessment requires combining visual reasoning over photos with policy language and regulatory constraints — a genuinely multimodal problem. The challenge was evaluating where Claude Vision reliably substitutes for adjuster judgment, and surfacing the cases where it diverges, without building on a live production system.
Outcomes
- Claude Vision damage classification from claimant-uploaded photos — type, severity, affected components
- Agentic pipeline: photo intake → damage classification → coverage determination → structured report
- Confidence scoring at each step with interpretable audit trail
- Pre-loaded demo claims with full AI assessments for rapid prototype evaluation
- In-memory store with production-ready interface for straightforward persistence swap