How I Approach Complex Systems
Fifteen years working on complex supply chains — and now on agentic AI systems that run them — have converged on a handful of convictions about what actually works.
Before optimizing anything, model the system's actual behavior — its variability, constraints, and feedback loops. Spreadsheets show averages; simulation and structured process modeling show reality. This applies whether the system is a spare-parts network or a multi-agent workflow.
Every project needs a quantifiable outcome tied to the real lever — cost, cycle time, service level, or decision quality. Generic metrics like 'efficiency' or 'productivity' hide the signal. Name the specific tradeoff being improved; make it measurable before you start.
Analysis that sits in a PowerPoint rarely changes anything. Build systems that translate insight into action — whether that's optimization software, a forecasting microservice, or an AI agent running a business process. Guardrails keep the system honest: agents act within human-defined policy, not around it.
Supply Chain Simulation
End-to-end discrete-event simulation models that capture stochastic behavior, multi-echelon inventory dynamics, and capacity constraints. Used for scenario planning, disruption response, and strategic network decisions.
Network & Cost Optimization
Mathematical optimization for logistics network design, facility location, routing, and duty/tariff analysis. Finding provably optimal solutions to complex multi-variable cost problems.
Demand Forecasting & Inventory Optimization
Forecasting models for intermittent and lumpy demand (the type traditional methods handle poorly), paired with inventory policies — safety stock, EOQ, ABC classification — that turn forecasts into replenishment decisions.
Agentic AI & Workflow Orchestration
Building multi-agent systems that run business processes — BPMN-designed workflows, MCP-based tool integrations, and Python microservices for domain-specific decisions. Guardrails keep the system honest: agents act within human-defined policy, not around it.