I built an L1 ticket deflector that knows when to stop: LangGraph + human-in-the-loop

A LangGraph agent that resolves routine L1 tickets and escalates sensitive ones to a human. Open-source, MIT, with an offline demo you can run on a Raspberry Pi.