video1.0<iframe src="https://www.loom.com/embed/0b9925b554a34fa0a8a8fee3e9388495" frameborder="0" width="1728" height="1296" webkitallowfullscreen mozallowfullscreen allowfullscreen></iframe>12961728Loomhttps://www.loom.com12961728https://cdn.loom.com/sessions/thumbnails/0b9925b554a34fa0a8a8fee3e9388495-0cf221d2bfda0e83.gif843.539This Loom presents Refund Pilot AI, a full stack customer support agent for e-commerce refund decisions. It simulates a workflow where the agent checks synthetic CRM and order data against a strict refund policy to approve, deny, escalate, or route to manual review, using deterministic policy validation to avoid customer manipulation. The frontend is built with Next.js and Tailwind, and the backend uses FastAPI with LangGraph orchestration and an Ollama LLM running locally. A demo covers multiple scenarios including damaged headphones leading to manual review typically taking 24 hours, smartwatch denial due to being outside a 30 day refund window, warranty handling, and cases requiring proof for manual verification. The project includes 15 synthetic CRM profiles and an admin dashboard with reasoning and notifications, plus voice input via speech to text.