I keep the scope narrow on purpose. Each project picks a specific piece of insurance work, builds something around it, and stays honest about where human review is still required.
Pulls useful information out of insurance documents - policy excerpts, submissions, loss runs - and answers structured questions with sources you can actually check.
Generates account briefs, renewal prep notes, and meeting materials from client inputs. Built around the kind of repetitive prep work that slows down every account team I've worked with.
A working framework for sorting AI use cases by risk level, setting review expectations, and building the controls you'd need before putting any of this in front of a regulated business.
No confidential or real client data. Everything uses mock or synthetic material.
Credibility in insurance means knowing what not to show. Clean repos, clear caveats, and no shortcuts on data privacy.