Insurance: Multi-Agent Claims Platform

The platform required three interconnected capabilities: an AI recommendation system for billing guidance, a cost estimation engine for commercial clients, and offline-first architecture for field operations. The recommendation system needed to generate job files with billing methodology trained on 2TB of historical data, using finetuned models with QLoRAs, a multi-agent system built on LangGraph with LLM-as-a-judge patterns to reduce hallucination, and human-in-the-loop workflows for low confidence outputs. RAG implementation enabled retrieval from historical records while data enrichment agents augmented outputs with current pricing using market, seasonal, and geographical rate data. The estimation system provided commercial clients with cost comparisons using historical job data rather than mathematical models. The PWA architecture supported field teams in low-connectivity environments with intelligent data reconciliation and optimistic state management.