Dynamic constraint-aware routing graph for cold-chain diagnostics logistics
Architected a real-time dispatch solver handling 18,000 daily diagnostic specimen transfers across strict temperature thresholds and variable clinic time windows.
Key Production Metric & Outcome
99.4% on-time sample arrival rate; cold-chain spoilage dropped by 31.8% in first quarter of production deployment.
A healthcare logistics provider operated a fleet of refrigerated courier vans transporting pathology specimens between 1,200 regional clinics and central processing laboratories. Schedules were originally calculated in nightly spreadsheet batches, leading to catastrophic delays when traffic or clinic delays occurred.
Problem & Hard Constraints
Every biological specimen possessed a strict 3-hour viable transit window under 2°C–8°C storage. A single delayed courier caused a cascading failure across downstream collection routes.
Architecture & Engineering Approach
We formulated route dispatch as an incremental constraint optimization problem over a PostGIS geospatial graph:
- Time-Window Graph Model: Explicit modeling of clinic operating hours, sample collection SLAs, courier capacity, and vehicle refrigeration telemetry.
- Incremental Solver Engine: Built on Google OR-Tools with customized heuristic search, re-solving updated routes in < 280ms when live GPS telemetry indicated a delay.
- Real-Time Snapshot Store: Redis state store providing dispatchers with an interactive topological map explaining why a particular collection stop was reassigned or reprioritized.
Chaos Engineering & Extreme Network Partition Handling
To ensure uninterrupted life-critical sample transfers during cellular blackouts in remote transport corridors, we implemented an offline-first vehicular consensus protocol. Couriers retain signed route state in encrypted local SQLite journals, allowing autonomous dynamic rerouting even when GPS or cloud connectivity is severed for over 45 minutes.