Logistics Sector
Engineering the OptiRoute Command Center
A reference build: streaming telematics into a routing model that re-optimises the whole fleet on a fixed cycle, and a dispatcher console that can draw ten thousand vehicles without dropping frames.
Design targets, not measured results
60sFull fleet re-optimisation cycle
3sTelematics ingest interval
WebGL10k+ vehicles on one canvas
Reactive Fleet Operations
The brief: a fleet in the tens of thousands, routed by hand, reacting to weather and traffic after they have already cost the day. Dispatchers were not short of data. They were short of a system that could act on it inside the window where acting still helps.
- warningManual routing, and fuel spend nobody could attribute.
- warningLive weather and traffic feeds arriving faster than anyone could read them.
Predictive Foundations
Prediction is the easy half. The hard half is a pipeline that can feed it and a console a dispatcher will actually trust at 6am.
route
Geospatial Analysis
Vector mapping and routing over a graph large enough that the algorithm choice, not the hardware, sets the response time.
insights
Model Scope
Historical traffic, weather and seasonal delay as features, plus an honest read on where a model beats a well-tuned heuristic, which is less often than it is sold.
speed
Real-time Telematics
A low-latency path for live vehicle diagnostics on a few-second interval, sized for the whole fleet reporting at once rather than for the average.
Build Sequence
Build Sequence
01 / 03
Phase 1: Data Pipeline Construction
Ingest, clean and standardise telematics from truck hardware of a dozen vintages. Most of the work in any AI project is here, and pretending otherwise is how AI projects fail.
02 / 03
Phase 2: Model Deployment
Inference serving with route scoring on the live path, versioned so a model can be rolled back exactly like code, because eventually one will need to be.
03 / 03
Phase 3: Dispatcher Console
A WebGL-accelerated command centre. Ten thousand moving markers is a rendering problem before it is a product problem, and it is solved at the canvas rather than in React.
The Predictive Mesh
Inference layered directly over streaming geospatial data, recalculating the optimal fleet state on a fixed cycle and dispatching micro-corrections. Fixed cycle rather than event-triggered on purpose: a system that re-routes drivers unpredictably is one drivers stop following.
Technology Stack
PythonTensorFlowReactMapbox GLAWS IoT
EDGE API GATEWAY
AUTH
ROUTING
DATA
DISTRIBUTED CLUSTER
Other reference builds
View All arrow_forwardRecognise your own system in this?
Then the next useful thing is forty-five minutes and a written brief on yours. Same format as this page, your constraints, and yours to keep either way.