CampusLoop

CampusLoop turns student move-out waste into affordable move-in kits. Students moving out upload a photo of an item. A vision agent grades its condition, color and defects, then prices it. Recalled models are blocked automatically. Students moving in set a budget, a minimum quality, colors and a room style, and a team of agents builds a matching kit. The agents only stop once the result is provably correct. Kit Builders (Nemotron Super) plan in parallel. A deterministic Quality Checker then runs 7 checks against the live inventory: availability, double-booking, recalls, budget, margin, coverage and quality. If a kit fails, the Coordinator sends the exact reasons back and the agents replan on their own. Anything they learn, such as a recalled model, is shared with every agent. After 3 failed attempts the Coordinator stops and hands the kit to a human. Nothing ships unverified. In short, the LLMs decide and the code verifies. Everything runs on Crusoe Managed Inference with Nemotron, DeepSeek V4 Flash and Gemma 4 vision. An admin "Agent control room" shows each request moving through the agents in real time: rejected, fixed and finally verified.