Apex
See the crisis before it hits
Apex is a supply-chain command platform that ingests real-time streams from 50+ carriers and port authorities, runs them through proprietary ML models, and surfaces 72-hour risk forecasts on an interactive 3D globe, turning raw noise into decisions.
Risk Radar
Year
2024
Client
Leading European logistics operator (confidential)
Role
Product design, system architecture, full-stack & ML modelling
Duration
9 months
Team
5 engineers, 1 data scientist, 1 designer
Category
Logistics & Supply Chain
// The context
2024 marked a peak in global route instability: Red Sea congestion, North Atlantic weather alerts, and bottlenecks at Northern European ports. The client was managing over 2,400 active shipments a day across four continents through disconnected tools and manually updated spreadsheets. The average window to react to a disruption was under six hours, often already too late.
// The challenge
The challenge wasn't collecting data. It was making it actionable in real time. Every carrier exposed different formats and cadences; weather and port data lived in incompatible silos; and the models had to score every active shipment every five minutes without tanking UI performance. Communicating risk with the clarity of a traffic light, without flattening the underlying complexity, was the hardest design constraint.
// The solution
We built a data lake on PostgreSQL with TimescaleDB for high-frequency time-series ingestion, orchestrating carrier feeds through Kafka for fault tolerance and controlled backpressure. The ML layer in Python with FastAPI exposes a scoring endpoint that combines ECMWF weather, port congestion indices, and carrier history into a single 72-hour risk score. The Next.js frontend consumes GraphQL and renders a WebGL globe with animated arcs encoding risk severity, volume, and direction in one 'god-mode' view.
Real-time 3D globe
A Three.js visualisation renders over 2,000 active routes, encoding risk, volume, and direction at 60fps on standard hardware.
72-hour risk forecasting
The ML model blends ECMWF weather, AIS port congestion, and carrier reliability into a single score refreshed every five minutes.
Unified data lake
A Kafka + TimescaleDB architecture normalises feeds from 50+ carriers into a single queryable schema, eliminating silos.
Proactive smart alerts
A tiered alert engine fires only above calibrated confidence thresholds, cutting alert noise by 74%.
Discovery & architecture
Four weeks shadowing operations teams to map data flows, real decision bottlenecks, and the ingestion architecture.
Data pipeline & ML
Built the data lake, normalised 50+ feeds, and iteratively trained the risk model on two years of historical disruptions.
Design & prototyping
Three rounds of high-fidelity prototyping, including A/B tests of the globe with real operators, before writing any WebGL.
Rollout & optimisation
Production launch with 40 pilot planners, then six weeks of alert-threshold tuning and model refinement.
Before Apex, our visibility ended at the warehouse door. Now we see the entire network on a single screen and know what's about to happen before it does; this isn't a new tool, it's a completely different way of running operations.
Marco Dellafiore, VP Operations, confidential client
// The impact
Apex turned a reactive operation into a predictive one: teams now intervene an average of 31 hours before a disruption turns critical. The client rolled the platform out to two additional business units within six months of launch.