Autonomous Replenishment Without the Bullwhip
Architected a production-grade demo eliminating supply chain distortion across five distributors using Temporal sagas.
Overview
I designed an autonomous continuous replenishment system to demonstrate how probabilistic forecasting and deterministic optimization eliminate the bullwhip effect. The architecture combines Temporal.io sagas for durable orchestration, TFT-based demand forecasting, and OR-Tools MILP for safety-stock allocation across a five-distributor vendor-managed inventory network. I delivered an interactive demo guide with persona-specific dashboards and allocation logic, giving stakeholders hands-on evidence that autonomous replenishment could run without human intervention. The system proved that ML predictions, when coupled with rigorous optimization and failure-aware workflow orchestration, could stabilize upstream orders despite downstream volatility.
Highlights
- 01
5-distributor VMI network with bullwhip elimination
- 02
Temporal sagas for durable replenishment workflows
- 03
OR-Tools MILP allocation with TFT forecasting
System Architecture
Probabilistic forecasting through MILP optimization to distributed replenishment via durable sagas.
Questions people ask
- How does this system eliminate the bullwhip effect?
- It combines TFT-based probabilistic forecasting with OR-Tools MILP optimization to stabilize upstream orders despite downstream volatility.
- What role does Temporal.io play in the architecture?
- Temporal sagas provide durable orchestration for replenishment workflows, ensuring the system handles failures without manual intervention.
- What technologies were used for the forecasting and allocation logic?
- I used Temporal Flow for workflows, TFT models for demand prediction, and OR-Tools MILP for safety-stock allocation.