Insurance claims copilot with long-term memory

The scenario

ClaimMate is a customer-facing insurance agent that can look up policies, open claims, request documents and report status through tool calls — and remembers each customer between conversations. Policy wording is answered through RAG, while personal context is stored in long-term memory, so returning customers never repeat themselves.

System architecture

High-level flow of ClaimMate, from input to output.

What you will learn

• Tool calling against a claims system
• Long-term vs. session memory for customer agents
• Combining policy RAG with personal context
• Safe handling of personal data in prompts

What you will build

• Claims API with policy, claim and document tools
• Agent with long-term memory
• Policy wording RAG
• Customer chat front end

Build plan

1. Mock claims system and tools
2. Agent with tool calling
3. Long-term memory
4. Policy RAG
5. Front end and testing

Tech stack