A real e-commerce return/refund process, run by a governed multi-agent AI system. Most "AI agent" demos are a chatbot with a nice prompt. ReturnGuard is the opposite: one company, one real operational process — reviewing every return/refund request — handed to a system of AI agents that is watched, bounded, and reversible. The agents auto-clear the obviously-legitimate returns and route anything risky, unclear, or expensive to a human, through a real fraud-review dashboard. Every decision is traceable to the exact model, prompt version, policy version, inputs, tokens, and cost that produced it. It's a portfolio project for Forward Deployed Engineer work — the emphasis is on the parts that are hard in the field: governed autonomy, full observability, per-agent identity, scenario-driven validation, a business case, and a clean handoff — not on "an agent that answers".

