video1.0<iframe src="https://www.loom.com/embed/d11c5fce945649c39c44e5acfcea53d5" frameborder="0" width="1662" height="1246" webkitallowfullscreen mozallowfullscreen allowfullscreen></iframe>12461662Loomhttps://www.loom.com12461662https://cdn.loom.com/sessions/thumbnails/d11c5fce945649c39c44e5acfcea53d5-d5a913df4a949760.gif186.432This Loom demonstrates an agent that auto-approves auto-insurance claims and escalates potential fraud patterns for manual review. The speaker, Pawan, explains that the system learns from each claim and, if it detects a pattern such as “closed in 40 minutes with no inspection” matching a known striping pattern at the same car park, it raises a nominated case. It then allows a reviewer to mark it as manually reviewed and critical by providing a reviewer name, triggering an agent flow in the next run. The agent is deployed in MongoDB, supports conversation-based access to policies, and includes guardrails that redact sensitive numbers and block system prompt related prompt injection attempts.