The strategic layer runs Monte Carlo Tree Search guided by a language model. Each node in the tree is one candidate restructuring; reward is a weighted blend of drive time, balance, cold-chain continuity, class affinity, and proximity. The agent reflects between iterations, and the best path becomes the recommended week.
Drop an FPSNila xlsx to run a fresh trace. The page streams MCTS events as they happen via Server-Sent Events. No data leaves the deployment.