When the ground
stops shaking, the math
begins.
An AI-augmented stochastic optimization platform that decides where to pre-position relief, how to route it, and when to re-plan — hedging across 10 demand scenarios simultaneously.
Stochastic Pre-Positioning
Two-stage MIP · Pyomo / Gurobi
Which warehouses to open, how many tons to stock — hedging across all demand scenarios at once.
- —Binary y[i] open decisions
- —Continuous inventory I[i]
- —Expected-value objective
CVRPTW Routing
Clarke-Wright + 2-opt · OR-Tools
Relief routes with capacity, time-windows and a carbon dimension. Lexicographic objective.
- —Multi-depot assignment
- —Priority-weighted demand
- —Carbon-aware objective
RL Re-Optimization
PPO · Stable-Baselines3
A learned policy decides WHEN to re-plan over a 72h episode — beating static and periodic baselines.
- —HOLD / RE-OPTIMIZE / ADD_TRUCK
- —State = demand + fleet + time
- —Reward = −α·unmet −β·cost −γ·CO₂
Demand Zones
0
3 critical priority
Candidate Sites
0
2080t total capacity
Scenarios
0
bootstrap + k-means reduced
Relief Fleet
6×5t
trucks · 45 km/h
Hedges uncertainty
Opens warehouses that perform well across all 10 scenarios — not just the expected case.
Solves in milliseconds
Subset enumeration + exact min-cost-flow per scenario. No waiting on solvers.
Beats the static baseline
99.9% coverage vs 91.8% static. 0.6t unmet vs 51.6t. Lower total cost.
§ 04 — Execute
Every minute is a life.
Run the full stochastic pipeline — pre-positioning, routing, and the learned re-optimization policy — and watch the map populate in real time.