Allstate Claim Supervisor Agent (Module 7)
π Description
This project simulates an intelligent claim supervisor named Anna who manages a team of agents to process auto insurance claims for Allstate. Built with LangChain, LangGraph, OpenAI GPT-4o-mini, and persistent memory, the system evaluates policy status, classifies damage severity, and determines payment outcomes. It features a modular, node-based structure that mirrors real insurance workflows.
Developed for the MSDS 442 course at Northwestern University, this project highlights how AI agents can be orchestrated into production-grade systems that reason over structured data and delegate tasks across teams β moving beyond chatbots to decision-making frameworks.
βΈ»
π§ Features
- Supervisor Agent: A human-like assistant named Anna delegates tasks to policy, claim, and damage agents.
- Memory Persistence: Each agent stores outputs to local files and vector memory for long-term tracking.
- Damage Classification: Uses structured prompts to label severity as Minor, Moderate, or Major.
- Vector Search Retrieval: Enables retrieval of past decisions and document references.
- Multi-Agent Workflow: Implements LangGraph with clear stages: input β verification β classification β decision.
- BPMN Compatibility: Designed to align with Business Process Modeling workflows and test cases.
βΈ»
π‘ Key Insight
Intelligent agents can be designed not just to chat β but to supervise. This project illustrates how modular LLM agents can handle real business logic, enabling automated workflows for high-volume, rules-based decision environments like insurance.
π View the source code on GitHub