Projects / Simulation · inventory policy · RL
Supply chain simulator
A small multi-echelon retail supply chain simulator. Factories, warehouses, shops and customers form a directed graph that shares one stochastic world with seasonality and disruptions. Each node runs its own replenishment policy, so ordering rules can be compared on identical random worlds.
A limited slice of the simulator. It may take a few seconds to wake up.
About the project
The simulator models a small retail supply chain as a directed graph. Factories produce, warehouses and shops hold stock and order from one or more suppliers, and customer demand is the only source of new money. All nodes share one stochastic world with regional supply and demand, seasonal cycles and disruptions, and deliveries arrive after a lead time set on each link.
Each node runs its own ordering policy. Every run is reproducible from a seed, so two policies can be compared on identical worlds and a difference in the results comes from the policy rather than from luck. The tuner and the reinforcement learning evaluation use the same pairing when they compare a candidate with the order-up-to baseline. Demand can also be replayed from real Walmart sales in the M5 dataset.
The live demo is a limited slice of the engine: three small graphs, the four textbook policies on one shop, and short runs. The rest of the project, including the reinforcement learning stack, the LLM world generator and M5 replay, runs from the repository. It is a demo project, written to be readable and easy to extend rather than production-grade.
Screenshots
